Publications resulting from research conducted using Delta appear here. Check back to see how the list of exciting discoveries made using Delta grows.
If you have a publication that should be listed here and isn’t, please share your success with us!
5005740
XHSH9DGT
1
nature
50
default
1693
https://delta.ncsa.illinois.edu/wp-content/plugins/zotpress/
%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A750%2C%22request_next%22%3A50%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%2254GJB5G2%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Mohapatra%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-25%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMohapatra%2C%20R.%2C%20Lancaster%2C%20L.%2C%20Fielding%2C%20D.%2C%20Quataert%2C%20E.%20%26amp%3B%20Bryan%2C%20G.%20L.%20How%20Magnetic%20Fields%20Regulate%20Cooling%20and%20Mixing%20in%20Turbulent%20Radiative%20Mixing%20Layers.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.29782%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.29782%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22How%20Magnetic%20Fields%20Regulate%20Cooling%20and%20Mixing%20in%20Turbulent%20Radiative%20Mixing%20Layers%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rajsekhar%22%2C%22lastName%22%3A%22Mohapatra%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lachlan%22%2C%22lastName%22%3A%22Lancaster%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Drummond%22%2C%22lastName%22%3A%22Fielding%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Eliot%22%2C%22lastName%22%3A%22Quataert%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Greg%20L.%22%2C%22lastName%22%3A%22Bryan%22%7D%5D%2C%22abstractNote%22%3A%22Turbulent%20radiative%20mixing%20layers%20%28TRMLs%29%20are%20expected%20wherever%20hot%20and%20cold%20gas%20move%20past%20one%20another%2C%20including%20in%20the%20solar%20corona%2C%20galactic%20winds%2C%20and%20cold%20filaments%20in%20galaxy%20clusters.%20These%20environments%20are%20often%20magnetized%2C%20but%20magnetic%20effects%20on%20mixing%20and%20cooling%20remain%20less%20well%20understood%20than%20in%20the%20hydrodynamic%20%28HD%29%20case.%20We%20present%20magnetohydrodynamic%20%28MHD%29%20simulations%20of%20TRMLs%20at%20resolutions%20up%20to%20%241024%20%5C%5Ctimes%202048%5E2%24%2C%20spanning%20fields%20aligned%20with%20and%20transverse%20to%20the%20shear%2C%20and%20polarity-reversing%20configurations%20in%20which%20oppositely%20directed%20fields%20form%20current%20sheets%20at%20the%20interface.%20Even%20initially%20weak%20hot-phase%20fields%2C%20with%20%24%5C%5Cmathcal%7BM%7D_%7B%5C%5Crm%20A%2Cshear%7D%5C%5Cequiv%20v_%7B%5C%5Crm%20shear%7D%5C%2Fv_%7B%5C%5Crm%20A%7D%5C%5Csim14%24%2C%20reduce%20the%20mass%20and%20enthalpy%20flux%20and%20radiative%20cooling%20rate%20by%20up%20to%20an%20order%20of%20magnitude%20relative%20to%20HD.%20Magnetic%20tension%20weakens%20turbulent%20motions%2C%20reducing%20both%20the%20diffusion%20of%20hot%20gas%20into%20the%20layer%20and%20the%20folding%20of%20the%20cooling%20surface.%20Transverse%20fields%20suppress%20cooling%20somewhat%20more%20strongly%20than%20shear-aligned%20ones%2C%20although%20a%20transverse%20field%20exerts%20no%20tension%20against%20the%20initial%20linear%20instability.%20Polarity%20reversal%20changes%20the%20morphology%20of%20the%20cooling%20gas%20without%20restoring%20HD-like%20mixing.%20The%20dependence%20on%20the%20Damk%5Cu00f6hler%20number%20is%20similar%20to%20the%20HD%20case%2C%20but%20the%20degree%20of%20suppression%20is%20dependent%20on%20the%20initial%20field%20orientation.%20The%20net%20cooling%20rate%20appears%20resolution-independent%20in%20HD%20but%20declines%20with%20resolution%20in%20MHD%2C%20and%20has%20not%20converged%2C%20so%20our%20suppression%20factors%20are%20lower%20limits.%20Magnetic%20fields%20therefore%20strongly%20regulate%20cooling%20and%20mixing%20in%20multiphase%20gas%2C%20and%20quantitative%20predictions%20require%20careful%20treatment%20of%20transport%20processes.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-25T00%3A58%3A55Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.29782%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.29782%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-27T19%3A28%3A00Z%22%7D%7D%2C%7B%22key%22%3A%22J4238H63%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Underwood%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-19%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BUnderwood%2C%20T.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Chronologic%3A%20Measuring%20Language%20Models%26%23x2019%3B%20Ability%20to%20Represent%20the%20Past.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.23178%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.23178%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Chronologic%3A%20Measuring%20Language%20Models%27%20Ability%20to%20Represent%20the%20Past%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ted%22%2C%22lastName%22%3A%22Underwood%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ziliang%22%2C%22lastName%22%3A%22Qiu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sarah%22%2C%22lastName%22%3A%22Griebel%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Laura%20K.%22%2C%22lastName%22%3A%22Nelson%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Edwin%22%2C%22lastName%22%3A%22Roland%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Wenyi%22%2C%22lastName%22%3A%22Shang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Matthew%22%2C%22lastName%22%3A%22Wilkens%22%7D%5D%2C%22abstractNote%22%3A%22Language%20models%20are%20appealing%20tools%20for%20research%20on%20the%20past.%20But%20to%20trust%20the%20evidence%20a%20model%20provides%2C%20researchers%20need%20to%20know%20whether%20its%20responses%20fit%20the%20period%20represented.%20Validation%20is%20challenging%2C%20because%20this%20is%20not%20a%20task%20living%20people%20ordinarily%20perform%2C%20and%20because%20many%20questions%20have%20multiple%20correct%20answers.%20We%20use%20historical%20texts%20to%20develop%20a%20benchmark%20for%20a%20model%26%23039%3Bs%20representation%20of%20English-language%20contexts%201831-1930%2C%20relying%20on%20pairwise%20comparisons%20to%20multiple%20ground%20truths%20and%20strong%20distractors%20to%20score%20the%20hardest%20questions%20in%20an%20appropriately%20graduated%20way.%20We%20find%20that%20generative%20tasks%20are%20harder%20than%20discriminative%20ones%3B%20in%20fact%2C%20reasoning%20models%20can%20typically%20discern%20the%20weakness%20of%20their%20own%20generated%20answers.%20While%20models%20pretrained%20exclusively%20on%20historical%20text%20lead%20the%20pack%20when%20evaluated%20by%20answer%20likelihood%2C%20they%20cannot%20compete%20with%20commercial%20models%20in%20free%20generation.%20None%20of%20the%20models%20we%20tested%20represent%20historical%20contexts%20in%20a%20fully%20persuasive%20way%20yet%2C%20but%20progress%20toward%20that%20goal%20is%20evident.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.23178%22%2C%22date%22%3A%222026-09-19%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.23178%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.23178%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-27T19%3A25%3A47Z%22%7D%7D%2C%7B%22key%22%3A%22LKXN3IHX%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Mohammadi%22%2C%22parsedDate%22%3A%222026-09-19%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMohammadi%2C%20M.%20Toscani-Fourier%20Distance%20on%20Probability%20Measures%3A%20Wasserstein%20Control%2C%20Topological%20Equivalence%20on%20Model%20Classes%2C%20and%20Duality.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.23163%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.23163%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Toscani-Fourier%20Distance%20on%20Probability%20Measures%3A%20Wasserstein%20Control%2C%20Topological%20Equivalence%20on%20Model%20Classes%2C%20and%20Duality%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mehrdad%22%2C%22lastName%22%3A%22Mohammadi%22%7D%5D%2C%22abstractNote%22%3A%22Comparing%20probability%20measures%20in%20machine%20learning%20trades%20transport%20geometry%20against%20computational%20cost%3A%20Wasserstein%20distances%20encode%20the%20geometry%20of%20%24%5C%5Cmathbb%20R%5Ed%24%20but%20require%20solving%20a%20transport%20problem%2C%20while%20kernel%20discrepancies%20are%20cheap%20to%20evaluate%20yet%20depend%20delicately%20on%20their%20test%20class.%20We%20study%20the%20Toscani--Fourier%20family%20%24%5C%5Cmathrm%20T_%7Bs%2Cp%7D%24%2C%20the%20weighted%20%24L%5Ep%24%20norm%20of%20the%20difference%20of%20two%20characteristic%20functions%2C%20as%20a%20continuous%20Fourier-side%20discrepancy%20on%20%24%5C%5Cmathbb%20R%5Ed%24.%20For%20%241%5C%5Cle%20p%26lt%3B%5C%5Cinfty%24%20we%20show%20that%20%24d%5C%2Fp%26lt%3Bs%26lt%3B1%2Bd%5C%2Fp%24%20is%20exactly%20the%20window%20in%20which%20%24%5C%5Cmathrm%20T_%7Bs%2Cp%7D%24%20is%20finite%20on%20%24%5C%5Cmathcal%20P_p%28%5C%5Cmathbb%20R%5Ed%29%24%2C%20both%20endpoints%20already%20failing%20for%20a%20pair%20of%20Dirac%20measures%2C%20and%20we%20establish%20the%20metric%2C%20embedding%2C%20and%20compactness%20structure%20of%20the%20resulting%20space%2C%20which%20we%20prove%20to%20be%20complete.%20Duality%20identifies%20%24%5C%5Cmathrm%20T_%7Bs%2Cp%7D%24%20as%20an%20integral%20probability%20metric%20over%20a%20homogeneous%20Fourier--Lebesgue%20ball%2C%20with%20an%20explicit%20extremizer%20when%20%241%26lt%3Bp%26lt%3B%5C%5Cinfty%24.%20We%20prove%20the%20global%20bound%20%24%5C%5Cmathrm%20T_%7Bs%2Cp%7D%5C%5Clesssim%20W_p%5E%7B%5C%5C%2Cs-d%5C%2Fp%7D%24%2C%20whose%20exponent%20is%20sharp%2C%20show%20that%20no%20global%20converse%20of%20any%20form%20can%20hold%2C%20and%20recover%20topological%20equivalence%20with%20%24W_p%24%20on%20bounded-support%20and%20uniform-tail%20classes%2C%20together%20with%20explicit%20reverse%20moduli%20on%20bounded-support%20classes%20that%20improve%20the%20imported%20energy-kernel%20exponent%20at%20%24p%3D2%24.%20There%2C%20%24%5C%5Cmathrm%20T_%7Bs%2C2%7D%24%20is%20a%20constant%20multiple%20of%20the%20classical%20energy%20distance%2C%20which%20yields%20an%20exact%20finite-sample%20identity%20for%20the%20mean%20of%20the%20empirical%20discrepancy%3B%20the%20numerical%20experiments%20are%20otherwise%20diagnostic.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.23163%22%2C%22date%22%3A%222026-09-19%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.23163%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.23163%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-27T19%3A24%3A01Z%22%7D%7D%2C%7B%22key%22%3A%22PBUWBMC7%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Li%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-19%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BLi%2C%20J.%2C%20Tian%2C%20J.%20%26amp%3B%20Watanabe%2C%20S.%20Speech%20Language%20Models%20for%20Full-Meeting%20Speaker%20Diarization%3A%20Capabilities%20and%20Limitations.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.23114%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.23114%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Speech%20Language%20Models%20for%20Full-Meeting%20Speaker%20Diarization%3A%20Capabilities%20and%20Limitations%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jialu%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jinchuan%22%2C%22lastName%22%3A%22Tian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22Recent%20advances%20in%20Speech%20Language%20Models%20%28SpeechLMs%29%2C%20which%20integrate%20large%20language%20models%20with%20speech%20foundation%20models%2C%20have%20enabled%20unified%20sequence%20modeling%20of%20speech%20processing%20tasks.%20However%2C%20many%20SpeechLM-based%20approaches%20to%20speaker%20diarization%20%28SD%29%20are%20tightly%20coupled%20with%20automatic%20speech%20recognition%20%28ASR%29%20and%20evaluated%20using%20word-level%20metrics%2C%20making%20it%20difficult%20to%20assess%20SD%20performance%20independent%20of%20ASR%20accuracy.%20In%20this%20work%2C%20we%20investigate%20ESPnet-SpeechLM%20as%20a%20token-based%20backbone%20for%20generating%20SD%20hypotheses%2C%20formulating%20SD%20as%20autoregressive%20generation%20of%20structured%20tokens%20conditioned%20on%20acoustic%20input.%20We%20systematically%20compare%20two%20output%20representations%3A%20an%20event-based%20representation%20that%20explicitly%20models%20speaker%20turn%20onset%20and%20offset%20timestamps%2C%20and%20a%20frame-based%20representation%20that%20predicts%20frame-level%20speaker%20activity.%20To%20provide%20structured%20conversational%20cues%2C%20we%20further%20incorporate%20auxiliary%20tasks%20including%20speech%20activity%20detection%2C%20overlapped%20speech%20detection%2C%20and%20speaker%20turn%20counting%20within%20the%20output%20sequence.%20Across%20multiple%20meeting%20datasets%2C%20we%20find%20that%20event-based%20representations%20produce%20more%20stable%20and%20consistent%20SD%20outputs%20than%20frame-based%20representations.%20Our%20analysis%20shows%20that%20outputs%20generated%20by%20SpeechLMs%20encode%20useful%20temporal%20SD%20structure%2C%20but%20full-meeting%20SD%20remains%20limited%20by%20recording-level%20speaker%20tracking%20and%20overlap-related%20misses.%20Explicit%20speaker-linking%20post-processing%20substantially%20reduces%20speaker%20confusion%2C%20suggesting%20that%20robust%20SpeechLM-based%20SD%20requires%20persistent%20speaker%20tracking%20and%20overlap-aware%20generation.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.23114%22%2C%22date%22%3A%222026-09-19%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.23114%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.23114%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-27T19%3A22%3A17Z%22%7D%7D%2C%7B%22key%22%3A%222CP8WFVS%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Gong%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-21%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BGong%2C%20J.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20SAM-V%3A%20Geometry-Aware%20Segment%20Anything%20for%20Multi-View%20Instance%20Segmentation.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.25490%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.25490%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22SAM-V%3A%20Geometry-Aware%20Segment%20Anything%20for%20Multi-View%20Instance%20Segmentation%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiangshan%22%2C%22lastName%22%3A%22Gong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yuqun%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qiqian%22%2C%22lastName%22%3A%22Fu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yao%22%2C%22lastName%22%3A%22Xiao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chuhang%22%2C%22lastName%22%3A%22Zou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shenlong%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Derek%22%2C%22lastName%22%3A%22Hoiem%22%7D%5D%2C%22abstractNote%22%3A%22Consistent%20multi-view%20object%20segmentation%20is%20critical%20for%203D%20perception%20and%20robotics%2C%20yet%20remains%20challenging%20under%20severe%20viewpoint%20and%20occlusion%20changes.%20Existing%20methods%20typically%20perform%203D%20instance%20segmentation%20on%20point%20clouds%20or%20rely%20on%20offline%202D%20mask-matching%20pipelines.%20However%2C%203D%20instance%20segmentation%20is%20limited%20by%20scarce%203D%20annotations%2C%20while%20offline%202D%20matching%20suffers%20from%20object%20identity%20ambiguity%20across%20frames.%20To%20leverage%20strong%202D%20and%203D%20priors%20jointly%2C%20we%20propose%20SAM-V%20%28Geometry-Aware%20Segment%20Anything%20for%20Multi-View%20Instance%20Segmentation%29.%20Instead%20of%20combining%20the%20two%20priors%20through%20post-hoc%20matching%2C%20SAM-V%20directly%20integrates%20features%20from%20a%20feed-forward%20geometry%20model%20%28VGGT%29%20into%20a%202D%20segmentation%20foundation%20model%20%28SAM%29%2C%20trained%20end-to-end%20for%20cross-view%20instance%20prediction.%20SAM-V%20introduces%20a%20prompt-fusion%20mechanism%20that%20enriches%20sparse%20SAM%20prompt%20tokens%20with%20view-specific%20camera%20tokens%20and%20local%20VGGT%20features%2C%20making%20the%20prompt%20representation%20both%20view-aware%20and%20spatially%20grounded%2C%20together%20with%20a%20mask%20decoder%20that%20attends%20to%20dense%202D%20and%203D%20features.%20By%20conditioning%20the%20mask%20decoding%20directly%20on%20multi-view%20geometry%2C%20SAM-V%20produces%20consistent%20multi-view%20segmentation%20of%20a%20prompted%20object%20in%20a%20single%20forward%20pass%20without%20offline%20mask%20matching%20or%20explicit%203D%20reconstruction.%20On%20the%20IGGT%203D%20tracking%20benchmark%2C%20where%20consistent%20instance%20identity%20across%20frames%20directly%20determines%20performance%2C%20SAM-V%20improves%20overall%20IoU%20by%205%20points%20and%20frame-level%20recall%20by%2012%20points%20on%20the%20ScanNet%2B%2B%20split%20over%20the%20state-of-the-art%20multi-view%20instance%20segmentation%20baseline%20and%20leads%20on%20all%20metrics%20in%20the%20zero-shot%20ScanNet%20split.%20Our%20code%20and%20pretrained%20models%20are%20available%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fgong208%5C%2FSAM-V.git.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.25490%22%2C%22date%22%3A%222026-09-21%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.25490%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.25490%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-27T19%3A18%3A45Z%22%7D%7D%2C%7B%22key%22%3A%22W8VEQGSS%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Kashaniyan%20and%20Jannesari%22%2C%22parsedDate%22%3A%222026-09-22%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BKashaniyan%2C%20M.%20%26amp%3B%20Jannesari%2C%20A.%20Sample%20Count%20Is%20Not%20Enough%3A%20Candidate-Generation%20Strategy%20Shapes%20the%20Energy%20and%20Performance%20of%20LLM%20Test-Time%20Scaling.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.19499%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.19499%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Sample%20Count%20Is%20Not%20Enough%3A%20Candidate-Generation%20Strategy%20Shapes%20the%20Energy%20and%20Performance%20of%20LLM%20Test-Time%20Scaling%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mobina%22%2C%22lastName%22%3A%22Kashaniyan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ali%22%2C%22lastName%22%3A%22Jannesari%22%7D%5D%2C%22abstractNote%22%3A%22Test-time%20scaling%20can%20improve%20large%20language%20model%20reasoning%20by%20generating%20and%20combining%20multiple%20candidate%20responses.%20In%20sampling-based%20methods%2C%20the%20inference%20budget%20is%20often%20described%20by%20the%20number%20of%20generated%20candidates%2C%20N.%20However%2C%20N%20tells%20us%20how%20many%20candidates%20are%20generated%2C%20not%20how%20they%20are%20executed.%20The%20same%20candidate%20budget%20can%20be%20produced%20in%20one%20batched%20generation%20call%20or%20split%20across%20several%20sequential%20calls%20with%20smaller%20batch%20sizes.%20We%20first%20study%20the%20effect%20of%20increasing%20N%20on%20reasoning%20accuracy%20using%20Phi-3-mini%20and%20Qwen2.5-1.5B%20on%20500%20GSM8K%20prompts.%20As%20expected%2C%20increasing%20N%20from%201%20to%208%20improves%20accuracy%20by%208.4%20percentage%20points%20for%20Phi-3-mini%20and%2018.4%20points%20for%20Qwen2.5-1.5B.%20However%2C%20accuracy%20alone%20does%20not%20show%20the%20systems%20cost%20of%20using%20a%20larger%20candidate%20budget.%20We%20therefore%20fix%20N%20%3D%208%20and%20compare%20four%20generation%20schedules%3A%201x8%2C%202x4%2C%204x2%2C%20and%208x1%2C%20where%20axb%20denotes%20a%20generation%20calls%20with%20b%20candidates%20per%20call.%20We%20measure%20latency%2C%20throughput%2C%20GPU-hours%2C%20and%20gross%20GPU-device%20energy%20while%20keeping%20the%20total%20candidate%20count%20fixed.%20On%20A100%20GPUs%2C%20eight%20serial%20calls%20use%204.64-4.86x%20as%20much%20gross%20GPU-device%20energy%20and%20have%205.77-6.12x%20the%20P95%20latency%20of%20one%20batched%20call%20with%20eight%20candidates.%20The%20same%20pattern%20appears%20across%20three%20independently%20scheduled%20A100%20nodes%20per%20model%20and%20in%20short-output%20SciQ%5C%2FV100%20experiments.%20These%20results%20show%20that%20candidate%20count%20alone%20is%20not%20enough%20to%20describe%20the%20systems%20cost%20of%20multi-candidate%20test-time%20scaling.%20When%20candidates%20are%20independent%20and%20memory%20allows%20it%2C%20fewer%20generation%20calls%20with%20larger%20batch%20sizes%20are%20more%20efficient.%20Evaluations%20should%20therefore%20report%20not%20only%20candidate%20count%20and%20accuracy%2C%20but%20also%20generation%20schedule%20and%20GPU-level%20systems%20metrics.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-22T00%3A47%3A30Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.19499%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.19499%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A54%3A51Z%22%7D%7D%2C%7B%22key%22%3A%22KDK8FXH5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hategan-Marandiuc%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-16%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHategan-Marandiuc%2C%20M.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Toward%20Autonomous%20Radio%20Follow-up%20of%20Multi-messenger%20Transients%20with%20RADAR%3A%20From%20Alert%20Parsing%20to%20Inference%20and%20Observation%20Scheduling.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.18233%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.18233%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Toward%20Autonomous%20Radio%20Follow-up%20of%20Multi-messenger%20Transients%20with%20RADAR%3A%20From%20Alert%20Parsing%20to%20Inference%20and%20Observation%20Scheduling%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mihael%22%2C%22lastName%22%3A%22Hategan-Marandiuc%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tanner%22%2C%22lastName%22%3A%22O%27Dwyer%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alessandra%22%2C%22lastName%22%3A%22Corsi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Eliu%22%2C%22lastName%22%3A%22Huerta%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zilinghan%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ian%20T.%22%2C%22lastName%22%3A%22Foster%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kyle%22%2C%22lastName%22%3A%22Chard%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ryan%22%2C%22lastName%22%3A%22Chard%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Amal%22%2C%22lastName%22%3A%22Gueroudji%22%7D%5D%2C%22abstractNote%22%3A%22Multi-messenger%20astronomy%20%28MMA%29%2C%20the%20joint%20study%20of%20cosmic%20sources%20through%20gravitational%20waves%20%28GWs%29%2C%20electromagnetic%20%28EM%29%20radiation%2C%20neutrinos%2C%20and%20cosmic%20rays%2C%20is%20rapidly%20reshaping%20time-domain%20astrophysics.%20Realizing%20the%20promise%20of%20MMA%20will%20require%20coordinating%20heterogeneous%20observing%20resources%20and%20automating%20the%20chain%20from%20alert%20to%20analysis%20to%20follow-up.%20RADAR%20%28Radio%20Afterglow%20Detection%20and%20AI-driven%20Response%29%20is%20a%20federated%2C%20privacy-enhancing%20framework%20for%20the%20radio%20follow-up%20of%20GW%20events%2C%20previously%20validated%20on%20GW170817.%20Here%2C%20we%20extend%20it%20along%20three%20axes.%20First%2C%20we%20benchmark%20three%20large%20language%20models%20%28LLMs%3B%20GPT-5.5%2C%20Claude-Opus-4.7%2C%20and%20Gemini-3.5-Flash%29%20against%20the%20GW170817%20radio%20light%20curve%20dataset.%20GPT-5.5%20attains%20the%20highest%20event-level%20%24F_1%24%20score%2C%20the%20harmonic%20mean%20of%20precision%20and%20recall%2C%20at%20%240.893%20%5C%5Cpm%200.010%24%2C%20and%20the%20highest%20GCN-level%20recall%2C%20%240.794%20%5C%5Cpm%200.013%24%2C%20improving%20on%20previous%20GPT-4.1%20results%20by%2016%5C%5C%25%20and%2010%5C%5C%25%2C%20respectively%2C%20while%20Claude-Opus-4.7%20achieves%20the%20highest%20precision%2C%20%240.978%20%5C%5Cpm%200.014%24.%20Second%2C%20we%20introduce%20concurrent%20likelihood%20evaluation%2C%20which%20speeds%20up%20the%20MCMC%20computation%20by%20a%20factor%20of%20%2440%5C%5Ctimes%24%20over%20our%20previous%20results.%20Third%2C%20we%20present%20an%20LLM-driven%20scheme%20that%20converts%20natural-language%20observing%20requests%20into%20submission-ready%20scheduling%20blocks%20for%20the%20Karl%20G.%20Jansky%20Very%20Large%20Array.%20Together%2C%20these%20developments%20advance%20RADAR%20toward%20a%20scalable%2C%20largely%20autonomous%20system%20for%20GW%20radio%20follow%20up.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.18233%22%2C%22date%22%3A%222026-09-16%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.18233%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.18233%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A52%3A49Z%22%7D%7D%2C%7B%22key%22%3A%22RQWWVFS5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Wu%20et%20al.%22%2C%22parsedDate%22%3A%222026-07-08%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BWu%2C%20Y.%2C%20Jeon%2C%20M.%2C%20Chang%2C%20J.-H.%20R.%2C%20Tuzel%2C%20O.%20%26amp%3B%20Tulsiani%2C%20S.%20RayRoPE%3A%20Projective%20Ray%20Positional%20Encoding%20for%20Multi-view%20Attention.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2601.15275%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2601.15275%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22RayRoPE%3A%20Projective%20Ray%20Positional%20Encoding%20for%20Multi-view%20Attention%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yu%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Minsik%22%2C%22lastName%22%3A%22Jeon%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jen-Hao%20Rick%22%2C%22lastName%22%3A%22Chang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Oncel%22%2C%22lastName%22%3A%22Tuzel%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shubham%22%2C%22lastName%22%3A%22Tulsiani%22%7D%5D%2C%22abstractNote%22%3A%22We%20study%20positional%20encodings%20for%20multi-view%20transformers%20that%20process%20tokens%20from%20a%20set%20of%20posed%20input%20images%2C%20and%20seek%20a%20mechanism%20that%20encodes%20patches%20uniquely%2C%20allows%20SE%283%29-invariant%20attention%20with%20multi-frequency%20similarity%2C%20and%20can%20adapt%20to%20the%20geometry%20of%20the%20underlying%203D%20scene.%20We%20find%20that%20prior%20%28absolute%20or%20relative%29%20encoding%20schemes%20for%20multi-view%20attention%20do%20not%20meet%20these%20desiderata%2C%20and%20present%20RayRoPE%20to%20address%20this%20gap.%20RayRoPE%20represents%20patch%20positions%20based%20on%20associated%20rays%20and%20computes%20query-frame%20projective%20coordinates%20to%20ensure%20SE%283%29%20invariance.%20To%20adapt%20to%20scene%20geometry%2C%20RayRoPE%20predicts%20%28without%20direct%20supervision%29%20a%20per-token%20depth%20to%20obtain%20its%20position%20along%20the%20corresponding%20ray%2C%20while%20also%20modeling%20uncertainty%20and%20analytically%20computing%20the%20expected%20positional%20encoding.%20We%20validate%20our%20method%20on%20the%20tasks%20of%20novel-view%20synthesis%2C%20stereo%20depth%20estimation%2C%20and%20feed-forward%203DGS%20reconstruction.%20While%20remaining%20efficient%2C%20RayRoPE%20consistently%20improves%20over%20alternate%20position%20encoding%20schemes%20%28e.g.%2C%2024%25%20relative%20improvement%20on%20LPIPS%20in%20RE10K%29.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-07-08T00%3A05%3A12Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2601.15275%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2601.15275%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A51%3A20Z%22%7D%7D%2C%7B%22key%22%3A%224SBDBZSD%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Trachu%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-16%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BTrachu%2C%20T.%2C%20Cornell%2C%20S.%2C%20Chen%2C%20W.%20%26amp%3B%20Watanabe%2C%20S.%20LACE%3A%20Layer-Wise%20Compression%20for%20Dynamic%20Frame%20Rate%20Codecs.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.17509%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.17509%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22LACE%3A%20Layer-Wise%20Compression%20for%20Dynamic%20Frame%20Rate%20Codecs%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Thanapat%22%2C%22lastName%22%3A%22Trachu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Samuele%22%2C%22lastName%22%3A%22Cornell%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22William%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22Neural%20audio%20codecs%20are%20a%20key%20component%20in%20speech%20language%20modeling.%20However%2C%20their%20high%20frame%20rates%20lead%20to%20long%20sequence%20lengths%2C%20increasing%20computational%20costs.%20Dynamic%20frame%20rate%20codecs%20mitigate%20this%20by%20reducing%20the%20effective%20frame%20rate%20using%20a%20compression%20step%20to%20merge%20multiple%20frames%20together.%20However%2C%20most%20prior%20methods%20either%20operate%20on%20single-codebook%20codecs%20or%20apply%20a%20single%20compression%20step%20before%20multi-layer%20quantization.%20This%20forces%20all%20quantization%20layers%20to%20share%20the%20same%20segmentation%20boundaries%2C%20despite%20the%20residual%20embeddings%20at%20different%20quantization%20layers%20exhibiting%20different%20rates%20of%20change%20over%20time.%20We%20propose%20LACE%20%28Layer-Adaptive%20Codec%20Encoding%29%2C%20a%20dynamic%20frame%20rate%20codec%20that%20applies%20an%20independent%20compression%20step%20at%20each%20quantization%20layer%2C%20enabling%20layer-specific%20segmentation%20boundaries.%20To%20use%20LACE%20tokens%20in%20downstream%20text-to-speech%20%28TTS%29%2C%20we%20further%20introduce%20union%20alignment%20and%20boundary%20anchor%20mechanisms%20to%20make%20durations%20consistent%20across%20layers%20while%20preserving%20compression%20benefits.%20Experiments%20on%20LibriTTS%20show%20that%20LACE%20offers%20a%20better%20rate-quality%20tradeoff%20than%20prior%20dynamic%20frame%20rate%20methods%20on%20the%20reconstruction%20task%20and%20improves%20TTS%20inference%20efficiency%20while%20maintaining%20competitive%20synthesis%20quality.%20Our%20code%20is%20released%20as%20part%20of%20the%20ESPnet3%20codec%20recipe.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-16T01%3A19%3A06Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.17509%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.17509%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A49%3A37Z%22%7D%7D%2C%7B%22key%22%3A%22X35YCUR6%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Gu%20et%20al.%22%2C%22parsedDate%22%3A%222026%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BGu%2C%20L.%2C%20Huang%2C%20H.-J.%2C%20Qadri%2C%20M.%2C%20Kaess%2C%20M.%20%26amp%3B%20Yuan%2C%20W.%20TouchAnything%3A%20Diffusion-Guided%203D%20Reconstruction%20from%20Sparse%20Robot%20Touches.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2604.08945%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2604.08945%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22TouchAnything%3A%20Diffusion-Guided%203D%20Reconstruction%20from%20Sparse%20Robot%20Touches%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Langzhe%22%2C%22lastName%22%3A%22Gu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hung-Jui%22%2C%22lastName%22%3A%22Huang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mohamad%22%2C%22lastName%22%3A%22Qadri%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Michael%22%2C%22lastName%22%3A%22Kaess%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Wenzhen%22%2C%22lastName%22%3A%22Yuan%22%7D%5D%2C%22abstractNote%22%3A%22Accurate%20object%20geometry%20estimation%20is%20essential%20for%20many%20downstream%20tasks%2C%20including%20robotic%20manipulation%20and%20physical%20interaction.%20Although%20vision%20is%20the%20dominant%20modality%20for%20shape%20perception%2C%20it%20becomes%20unreliable%20under%20occlusions%20or%20challenging%20lighting%20conditions.%20In%20such%20scenarios%2C%20tactile%20sensing%20provides%20direct%20geometric%20information%20through%20physical%20contact.%20However%2C%20reconstructing%20global%203D%20geometry%20from%20sparse%20local%20touches%20alone%20is%20fundamentally%20underconstrained.%20We%20present%20TouchAnything%2C%20a%20framework%20that%20leverages%20a%20pretrained%20large-scale%202D%20vision%20diffusion%20model%20as%20a%20semantic%20and%20geometric%20prior%20for%203D%20reconstruction%20from%20sparse%20tactile%20measurements.%20Unlike%20prior%20work%20that%20trains%20category-specific%20reconstruction%20networks%20or%20learns%20diffusion%20models%20directly%20from%20tactile%20data%2C%20we%20transfer%20the%20geometric%20knowledge%20encoded%20in%20pretrained%20visual%20diffusion%20models%20to%20the%20tactile%20domain.%20Given%20sparse%20contact%20constraints%20and%20a%20coarse%20class-level%20description%20of%20the%20object%2C%20we%20formulate%20reconstruction%20as%20an%20optimization%20problem%20that%20enforces%20tactile%20consistency%20while%20guiding%20solutions%20toward%20shapes%20consistent%20with%20the%20diffusion%20prior.%20Our%20method%20reconstructs%20accurate%20geometries%20from%20only%20a%20few%20touches%2C%20outperforms%20existing%20baselines%2C%20and%20enables%20open-world%203D%20reconstruction%20of%20previously%20unseen%20object%20instances.%20Our%20project%20page%20is%20https%3A%5C%2F%5C%2Fgrange007.github.io%5C%2Ftouchanything%20.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2604.08945%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2604.08945%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A48%3A28Z%22%7D%7D%2C%7B%22key%22%3A%2268MPRYE9%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Zhu%20et%20al.%22%2C%22parsedDate%22%3A%222025%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BZhu%2C%20Z.%2C%20Gong%2C%20Y.%2C%20Xiao%2C%20Y.%2C%20Liu%2C%20Y.%20%26amp%3B%20Hoiem%2C%20D.%20How%20to%20Teach%20Large%20Multimodal%20Models%20New%20Skills.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2510.08564%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2510.08564%26lt%3B%5C%2Fa%26gt%3B%20%282025%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22How%20to%20Teach%20Large%20Multimodal%20Models%20New%20Skills%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhen%22%2C%22lastName%22%3A%22Zhu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yiming%22%2C%22lastName%22%3A%22Gong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yao%22%2C%22lastName%22%3A%22Xiao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yaoyao%22%2C%22lastName%22%3A%22Liu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Derek%22%2C%22lastName%22%3A%22Hoiem%22%7D%5D%2C%22abstractNote%22%3A%22How%20can%20we%20teach%20large%20multimodal%20models%20%28LMMs%29%20new%20skills%20without%20erasing%20prior%20abilities%3F%20We%20study%20sequential%20fine-tuning%20on%20five%20target%20skills%20while%20monitoring%20general%20ability%20on%20eight%20held-out%20benchmarks%20across%20three%20model%20families.%20Surprisingly%2C%20we%20find%20that%20performance%20lost%20on%20held-out%20tasks%20after%20fine-tuning%20on%20one%20skill%20can%20partly%20recover%20when%20the%20model%20is%20subsequently%20tuned%20on%20a%20different%20skill.%20We%20trace%20this%20behavior%20to%20a%20measurable%20shift%20in%20the%20output%20token%20distribution%2C%20manifested%20through%20a%20simple%20counting-bias%20probe%20that%20shows%20the%20shift%20co-varies%20with%20forgetting.%20Guided%20by%20this%20insight%2C%20we%20identify%20two%20simple%2C%20robust%20tuning%20recipes%20that%20learn%20strongly%20while%20limiting%20drift%3A%20%28i%29%20updating%20only%20the%20self-attention%20projection%20layers%20%28SA%20Proj.%2C%20%24%5Cu0394%24%20learning%20%2B24.9%20%5C%2F%20%24%5Cu0394%24%20held-out%20forgetting%20-0.6%29%2C%20and%20%28ii%29%20updating%20only%20the%20MLP%20Gate%26amp%3BUp%20while%20freezing%20the%20Down%20projection%20%28%2B30.5%20%5C%2F%20-2.1%29.%20Both%20substantially%20outperform%20full-LLM%20tuning%20%28%2B31.8%20%5C%2F%20-23.3%29%20in%20the%20learning-forgetting%20trade-off.%20We%20also%20compare%20against%20common%20forgetting%20mitigation%20methods%3A%20Learning%20without%20Forgetting%20%28LwF%29%2C%20LoRA%2C%20Mixture-of-Experts%2C%20and%20weight-space%20interpolation%20%28WiSE-FT%29%2C%20and%20find%20that%20our%20selective%20tuning%20recipes%20match%20or%20exceed%20their%20learning-stability%20balance%20while%20remaining%20simpler%2C%20requiring%20no%20replay%2C%20auxiliary%20parameters%2C%20or%20per-stage%20tuning.%20These%20results%20hold%20across%20LLaVA-OneVision%2C%20LLaVA-NeXT%2C%20and%20Qwen2.5-VL%2C%20confirming%20that%20the%20key%20to%20teaching%20LMMs%20new%20skills%20without%20forgetting%20lies%20in%20controlling%20output%20distribution%20shift%20by%20choosing%20which%20components%20to%20tune.%20Code%20will%20be%20made%20available.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222025%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2510.08564%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2510.08564%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A44%3A51Z%22%7D%7D%2C%7B%22key%22%3A%22ZUN4TRGX%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Tang%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-11%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BTang%2C%20B.%2C%20Spalding%2C%20Z.%20%26amp%3B%20Cogan%2C%20G.%20B.%20Pretraining%20for%20Sample-Efficient%20Neural%20Interfaces.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.13507%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.13507%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Pretraining%20for%20Sample-Efficient%20Neural%20Interfaces%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ben%22%2C%22lastName%22%3A%22Tang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zachary%22%2C%22lastName%22%3A%22Spalding%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Gregory%20B.%22%2C%22lastName%22%3A%22Cogan%22%7D%5D%2C%22abstractNote%22%3A%22Brain-computer%20interfaces%20%28BCIs%29%20decode%20neural%20activity%20to%20restore%20lost%20function.%20Typically%2C%20training%20a%20high-performance%20neural%20decoder%20requires%20a%20large%20labeled%20dataset%20to%20be%20collected%20from%20every%20new%20subject.%20One%20way%20to%20reduce%20the%20labeled%20data%20cost%20is%20self-supervised%20pretraining%2C%20which%20learns%20general%20neural%20representations%20from%20unlabeled%20recordings%20that%20accumulate%20across%20subjects.%20However%2C%20for%20intracranial%20electroencephalography%20%28iEEG%29%20recordings%2C%20self-supervised%20learning%20has%20been%20challenging%20due%20to%20differences%20in%20contact%20placement%20and%20neuroanatomy%20between%20subjects.%20We%20propose%20MAPA%2C%20an%20otherwise%20vanilla%20masked%20autoencoder%20with%20two%20spatial%20encodings%2C%20an%20anatomical%20region%20embedding%20and%20a%20relative%20positional%20encoding%2C%20which%20together%20enable%20it%20to%20learn%20neural%20representations%20that%20transfer%20to%20unseen%20subjects%20and%20across%20various%20tasks.%20MAPA%20sets%20a%20new%20state%20of%20the%20art%20across%20all%20three%20regimes%20of%20the%20Neuroprobe%20benchmark%20without%20fine-tuning%3A%20within-session%2C%20cross-session%2C%20and%20cross-subject.%20In%20the%20cross-subject%20regime%2C%20a%20linear%20probe%20on%20MAPA%26%23039%3Bs%20features%20needs%20only%20%24%7B%5C%5Csim%7D164%24%20labeled%20trials%20to%20reach%20the%20accuracy%20that%20takes%203%2C500%20without%20pretraining.%20Our%20results%20show%20that%20self-supervised%20pretraining%20can%20scale%20across%20heterogeneous%20iEEG%20recordings%20and%20reduce%20the%20labeled%20data%20needed%20for%20accurate%20decoding%20in%20new%20subjects.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.13507%22%2C%22date%22%3A%222026-09-11%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.13507%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.13507%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A38%3A56Z%22%7D%7D%2C%7B%22key%22%3A%22DIZTG3N6%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Trepka%22%2C%22parsedDate%22%3A%222026%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BTrepka%2C%20E.%20An%20Evaluation%20of%20the%20Two-Dimensional%20Gabor%20Model%20of%20First-Layer%20Filters%20in%20Convolutional%20Neural%20Networks.%20in%20%26lt%3Bi%26gt%3BECCV%202026%20Workshop%26lt%3B%5C%2Fi%26gt%3B%20%28Malmo%2C%20Sweden%2C%202026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22An%20Evaluation%20of%20the%20Two-Dimensional%20Gabor%20Model%20of%20First-Layer%20Filters%20in%20Convolutional%20Neural%20Networks%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ethan%22%2C%22lastName%22%3A%22Trepka%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22proceedingsTitle%22%3A%22ECCV%202026%20Workshop%22%2C%22conferenceName%22%3A%22Human-inspired%20Computer%20Vision%22%2C%22date%22%3A%222026%22%2C%22eventPlace%22%3A%22%22%2C%22DOI%22%3A%22%22%2C%22ISBN%22%3A%22%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fopenreview.net%5C%2Fpdf%3Fid%3Dojn1rg914P%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A37%3A56Z%22%7D%7D%2C%7B%22key%22%3A%22CNL6PF3D%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Syfullah%20and%20Khalil%22%2C%22parsedDate%22%3A%222026-09-15%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BSyfullah%2C%20M.%20K.%20%26amp%3B%20Khalil%2C%20A.%20A.%20LLM-Based%20Schema-Aware%20Split%20Learning%20for%20Privacy-Preserving%20Mental%20Distress%20Prediction%20Across%20Heterogeneous%20Surveys.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.15871%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.15871%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22LLM-Based%20Schema-Aware%20Split%20Learning%20for%20Privacy-Preserving%20Mental%20Distress%20Prediction%20Across%20Heterogeneous%20Surveys%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Md%20Khalid%22%2C%22lastName%22%3A%22Syfullah%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alvi%20Ataur%22%2C%22lastName%22%3A%22Khalil%22%7D%5D%2C%22abstractNote%22%3A%22Rising%20societal%20and%20lifestyle%20complexity%20has%20been%20linked%20to%20a%20growing%20prevalence%20of%20mental%20distress%20worldwide.%20Educational%20institutions%2C%20workplaces%2C%20clinics%2C%20etc.%20collect%20large%20volumes%20of%20mental%20health%20survey%20data%20to%20understand%20and%20reduce%20this%20burden.%20Collaborative%20analysis%20of%20such%20data%20could%20yield%20effective%20generalizable%20predictive%20models.%20Privacy%20constraints%20and%20varied%20survey%20designs%20%28i.e.%2C%20different%20questions%2C%20scales%2C%20and%20formats%29%20hinder%20direct%20integration.%20We%20propose%20a%20schema-aware%20split%20learning%20%28SL%29%20framework%20that%20preserves%20privacy%2C%20using%20a%20large%20language%20model%20%28LLM%29%20as%20a%20shared%20semantic%20encoder%20to%20harmonize%20heterogeneous%20survey%20schemas%20across%20institutions.%20We%20serialize%20each%20survey%20record%20into%20a%20natural-language%20description%2C%20unifying%20disparate%20survey%20schemas%20into%20a%20common%20format.%20The%20LLM%20is%20fine-tuned%20for%20mental%20distress%20assessment%20via%20Low-Rank%20Adaptation%20%28LoRA%29%20and%20partitioned%20across%20client%20and%20server.%20Clients%20retain%20the%20raw%20survey%20responses%20locally%20and%20run%20only%20a%20lightweight%20front-end%2C%20so%20original%20records%20never%20leave%20the%20institution%20that%20collected%20them.%20The%20resource-intensive%20backbone%20runs%20on%20the%20server%2C%20minimizing%20client-side%20computation.%20Using%20LLaMA-3.2-3B-Instruct%2C%20the%20framework%20attains%20an%20average%20ANLS%20of%200.708%20with%20only%202%2C000%20training%20samples%2C%20surpasses%20federated%20learning%20%28FL%29%20in%20eight%20of%20nine%20settings%2C%20and%20cuts%20per-client%20computation%20by%20three%20orders%20of%20magnitude%2C%20while%20generalizing%20to%20unseen%20datasets.%20Overall%2C%20it%20enables%20accurate%2C%20privacy-preserving%2C%20and%20resource-efficient%20collaborative%20learning%20from%20heterogeneous%20mental%20health%20survey%20data.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-15T02%3A19%3A58Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.15871%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.15871%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A33%3A10Z%22%7D%7D%2C%7B%22key%22%3A%22BRGJMWM5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Oza%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-18%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BOza%2C%20P.%2C%20Fung%2C%20V.%20%26amp%3B%20Hu%2C%20G.%20Reaction-Condition-%20and%20Ensemble-Aware%20Screening%20of%20Dual-Atom%20ORR%20Catalysts%20via%20Physics-Informed%20Machine%20Learning.%20%26lt%3Bi%26gt%3BACS%20Catalysis%26lt%3B%5C%2Fi%26gt%3B%20%26lt%3Bb%26gt%3B16%26lt%3B%5C%2Fb%26gt%3B%2C%2018174%26%23x2013%3B18182%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Reaction-Condition-%20and%20Ensemble-Aware%20Screening%20of%20Dual-Atom%20ORR%20Catalysts%20via%20Physics-Informed%20Machine%20Learning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Prajeet%22%2C%22lastName%22%3A%22Oza%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Victor%22%2C%22lastName%22%3A%22Fung%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Guoxiang%22%2C%22lastName%22%3A%22Hu%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Metal%5Cu2013nitrogen%5Cu2013carbon%20dual-atom%20catalysts%20%28M1%5C%2FM2%5Cu2013N%5Cu2013C%20DACs%29%20have%20emerged%20as%20promising%20alternatives%20to%20Pt-based%20catalysts%20for%20the%20oxygen%20reduction%20reaction%20%28ORR%29%2C%20yet%20their%20rational%20discovery%20is%20hindered%20by%20an%20enormous%20chemical%20and%20structural%20design%20space.%20Here%2C%20we%20develop%20a%20physics-informed%20machine%20learning%20%28ML%29%20framework%20integrated%20with%20high-throughput%20density%20functional%20theory%20%28DFT%29%20to%20systematically%20screen%2022%2C599%20M1%5C%2FM2%5Cu2013N%5Cu2013C%20DAC%20structures%20spanning%20729%20transition-metal%20pairs%20and%2031%20structural%20configurations.%20Rather%20than%20relying%20on%20idealized%20bare-surface%20models%2C%20we%20construct%20voltage-dependent%20ab%20initio%20thermodynamic%20phase%20diagrams%20to%20identify%20realistic%20active-site%20structures%20and%20ORR%20limiting%20potentials%20under%20electrochemical%20operating%20conditions.%20To%20accelerate%20screening%2C%20we%20train%20equivariant%20transformer%20graph%20neural%20networks%20on%20DFT-generated%20adsorption%20energetics%2C%20achieving%20high%20predictive%20accuracy%20with%20mean%20absolute%20errors%20as%20low%20as%200.015%20eV%20for%20adsorption%20energies%20and%200.022%20V%20for%20derived%20ORR%20limiting%20potentials.%20We%20further%20introduce%20an%20experimentally%20relevant%20descriptor%2C%20the%20percentage%20of%20catalytically%20active%20structural%20configurations%20for%20each%20metal%20pair%2C%20which%20captures%20the%20ensemble%20nature%20of%20experimentally%20synthesized%20DACs%20and%20provides%20more%20reliable%20catalyst%20ranking%20than%20conventional%20single-configuration%20approaches.%20The%20framework%20identifies%203431%20DAC%20structures%20with%20predicted%20ORR%20limiting%20potentials%20exceeding%200.8%20V%20and%20reveals%20that%20only%20a%20small%20subset%20of%20metal%20pairs%20exhibits%20consistently%20high%20activity%20across%20configurations.%20Several%20top-performing%20candidates%2C%20including%20Co%5C%2FCr%2C%20Co%5C%2FAg%2C%20Co%5C%2FRu%2C%20Co%5C%2FIr%2C%20and%20Co%5C%2FZn%2C%20are%20validated%20by%20additional%20DFT%20calculations%20and%20show%20strong%20agreement%20with%20available%20experimental%20trends.%20In%20addition%2C%20stability%20screening%20uncovers%20multiple%20DACs%20predicted%20to%20possess%20both%20higher%20ORR%20activity%20and%20greater%20thermodynamic%20stability%20than%20benchmark%20Fe%5C%2FCo%5Cu2013N%5Cu2013C%20catalysts.%20This%20work%20establishes%20a%20scalable%20ML-assisted%20paradigm%20for%20realistic%20electrocatalyst%20screening%20and%20provides%20design%20principles%20for%20the%20discovery%20of%20next-generation%20dual-atom%20ORR%20catalysts.%22%2C%22date%22%3A%222026-09-18%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1021%5C%2Facscatal.6c04252%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fpubs.acs.org%5C%2Faccacs%5C%2Farticle%5C%2F16%5C%2F18%5C%2F18174%5C%2F5427308%5C%2FReaction-Condition-and-Ensemble-Aware-Screening-of%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%222155-5435%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A31%3A36Z%22%7D%7D%2C%7B%22key%22%3A%22NI23T6FI%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Ge%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-11%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BGe%2C%20Y.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20MindTopo%3A%20Can%20Foundation%20Models%20Reason%20in%20Topological%20Space%3F%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.11900%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.11900%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22MindTopo%3A%20Can%20Foundation%20Models%20Reason%20in%20Topological%20Space%3F%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yunfei%22%2C%22lastName%22%3A%22Ge%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Anbang%22%2C%22lastName%22%3A%22Liu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qineng%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Johnalbert%22%2C%22lastName%22%3A%22Garnica%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jianwen%22%2C%22lastName%22%3A%22Lyu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zihan%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Reuben%22%2C%22lastName%22%3A%22Tan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jianfeng%22%2C%22lastName%22%3A%22Gao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ruohan%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yining%22%2C%22lastName%22%3A%22Hong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiajun%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Manling%22%2C%22lastName%22%3A%22Li%22%7D%5D%2C%22abstractNote%22%3A%22Spatial%20reasoning%20depends%20not%20only%20on%20metric%20properties%20such%20as%20distance%2C%20angle%2C%20and%20shape%2C%20but%20also%20on%20topological%20relations%20that%20remain%20invariant%20under%20continuous%20deformation.%20Cognitive%20science%20identifies%20these%20relations%20as%20foundational%20to%20spatial%20understanding%2C%20yet%20foundation-model%20evaluations%20largely%20focus%20on%20metric%20or%20viewpoint-dependent%20relations.%20We%20introduce%20MindTopo%2C%20a%20benchmark%20of%20topological%20intuition%20across%20five%20properties%20grounded%20in%20cognitive%20science%20and%20formal%20topology%3A%20continuity%2C%20separation%2C%20order%2C%20enclosure%2C%20and%20knots.%20MindTopo%20evaluates%20each%20property%20at%20two%20cognitive%20levels.%20Reasoning%20asks%20a%20model%20to%20identify%20topological%20relations%20or%20infer%20how%20they%20change.%20Planning%20instantiates%20a%20foundation%20model%20as%20a%20closed-loop%20agent%20whose%20policy%20selects%20environment%20actions.%20MindTopo%20contains%2011%2C030%20instances%20across%2013%20procedurally%20generated%20task%20types%20with%20controllable%20difficulty.%20We%20benchmark%2014%20MLLMs%20and%20study%20agent%20configurations%20augmented%20with%20image%20and%20video%20generation%2C%20including%203%20video%20generative%20models%20in%20planning%20settings.%20Every%20MLLM%20performs%20better%20on%20reasoning%20than%20on%20planning%2C%20and%20the%20best-performing%20model%20remains%20far%20below%20observed%20human%20performance.%20On%20Qwen3-VL-2B-Instruct%2C%20supervised%20fine-tuning%20and%20reinforcement%20learning%20improve%20reasoning%20more%20than%20planning.%20Generated%20observations%20retain%20local%20cues%20and%20reach%20plausible%20endpoints%2C%20but%20audited%20rollouts%20do%20not%20reliably%20follow%20environment%20dynamics%20or%20preserve%20topology%20across%20transitions.%20Our%20website%20is%20at%20https%3A%5C%2F%5C%2Fmind-topo.github.io%5C%2F%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-11T01%3A11%3A50Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.11900%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.11900%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A27%3A49Z%22%7D%7D%2C%7B%22key%22%3A%22KIYGP42Z%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Kwark%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-09%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BKwark%2C%20D.%20H.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Seamless%20Whole%20Slide%20Label-Free%20Virtual%20Staining.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.10914%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.10914%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Seamless%20Whole%20Slide%20Label-Free%20Virtual%20Staining%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dou%20Hoon%22%2C%22lastName%22%3A%22Kwark%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kianoush%22%2C%22lastName%22%3A%22Falahkheirkhah%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ji-hun%22%2C%22lastName%22%3A%22Oh%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shirui%22%2C%22lastName%22%3A%22Luo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Volodymyr%22%2C%22lastName%22%3A%22Kindratenko%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rohit%22%2C%22lastName%22%3A%22Bhargava%22%7D%5D%2C%22abstractNote%22%3A%22Label-free%20virtual%20staining%20offers%20a%20compelling%2C%20non-destructive%20alternative%20to%20standard%20histopathology%3B%20however%2C%20its%20clinical%20adoption%20is%20hindered%20by%20the%20computational%20bottlenecks%20inherent%20to%20processing%20gigapixel%20Whole%20Slide%20Images%20%28WSIs%29.%20Current%20deep%20learning%20approaches%20require%20patch-based%20inference%20to%20avoid%20memory%20constraints%2C%20which%20disrupts%20global%20tissue%20continuity%20and%20introduces%20tiling%20artifacts--displaying%20visible%20seams%20and%20color%20shifts.%20To%20address%20this%2C%20we%20introduce%20the%20Consistency%20Memory%20Bank%20%28COMB%29%2C%20a%20novel%20label-free%20virtual%20staining%20framework%20that%20enforces%20spatial%20and%20channel%20consistency%20across%20tiles%20without%20memory%20bottlenecks.%20COMB%20decouples%20context%20storage%20from%20computation%2C%20utilizing%20a%20dynamic%20retrieval%20mechanism%20to%20fetch%20feature%20representations%20from%20adjacent%20tiles.%20This%20enables%20a%20retrieval-based%20context%20integration%20strategy%20that%20adopts%20local%20padding%20to%20resolve%20spatial%20discontinuities%20and%20neighbor-aware%20channel%20attention%20to%20stabilize%20statistical%20drift.%20Further%20optimized%20with%20a%20sliding%20window%20schedule%20to%20ensure%20minimal%20memory%20overhead%2C%20our%20method%20demonstrates%20superior%20performance%20over%20state-of-the-art%20baselines%2C%20achieving%20significant%20improvements%20in%20both%20perceptual%20fidelity%20and%20tiling%20consistency%2C%20while%20suggesting%20its%20downstream%20utility%20in%20tumor%20segmentation.%20Code%20is%20available%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fdou0000%5C%2FCOMB.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.10914%22%2C%22date%22%3A%222026-09-09%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.10914%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.10914%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A22%3A52Z%22%7D%7D%2C%7B%22key%22%3A%228E6LKHZU%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22G%5Cu00fcne%5Cu015f%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-09%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BG%26%23xFC%3Bne%26%23x15F%3B%2C%20M.%20H.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Signature%20of%20Short-Range%20Order%20in%20Static%20Response%20of%20the%20Three-Dimensional%20Electron%20Gas.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.10006%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2609.10006%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Signature%20of%20Short-Range%20Order%20in%20Static%20Response%20of%20the%20Three-Dimensional%20Electron%20Gas%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Muhammed%20H.%22%2C%22lastName%22%3A%22G%5Cu00fcne%5Cu015f%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yubo%22%2C%22lastName%22%3A%22Yang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Vitaly%22%2C%22lastName%22%3A%22Gorelov%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Miguel%20A.%22%2C%22lastName%22%3A%22Morales%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Matteo%22%2C%22lastName%22%3A%22Gatti%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lucia%22%2C%22lastName%22%3A%22Reining%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shiwei%22%2C%22lastName%22%3A%22Zhang%22%7D%5D%2C%22abstractNote%22%3A%22The%20three-dimensional%20electron%20gas%20is%20a%20fundamental%20model%20in%20condensed%20matter%20physics%20and%20quantum%20chemistry%2C%20and%20the%20exchange-correlation%20energy%20derived%20from%20it%20serves%20as%20the%20starting%20point%20of%20%5C%5Ctextit%7Bab%20initio%7D%20computations%20of%20materials.%20However%2C%20the%20wave-vector-dependent%20response%2C%20and%20hence%20the%20static%20local%20field%20factor%20%24G%28q%29%24%2C%20has%20remained%20without%20accurate%20ground-state%20benchmark%20in%20the%20strongly%20coupled%20regime%20for%20three%20decades.%20Using%20diffusion%20Monte%20Carlo%2C%20we%20calculate%20%24G%28q%29%24%20and%20the%20static%20density-density%20response%20function%20across%20the%20liquid%20phase%20and%20find%20a%20pronounced%20structure%20in%20%24G%28q%29%24%20at%20intermediate%20wave%20vectors%2C%20already%20visible%20at%20metallic%20densities%20and%20growing%20with%20increasing%20interaction%20strength.%20We%20identify%20it%20as%20a%20fingerprint%20of%20short-range%20order%20by%20showing%20that%20it%20is%20required%20to%20reproduce%20the%20static%20structure%20factor.%20Our%20parametrization%2C%20valid%20in%20the%20entire%20liquid%20phase%2C%20predicts%20a%20low-energy%20resonant%20mode%20inside%20the%20particle-hole%20continuum.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2609.10006%22%2C%22date%22%3A%222026-09-09%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2609.10006%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.10006%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A21%3A46Z%22%7D%7D%2C%7B%22key%22%3A%22KJXP7LL5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Marx%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-10%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMarx%2C%20D.%20C.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Structural%20basis%20of%20active%20state%20coupling%20of%20metabotropic%20glutamate%20receptor%208%20to%20beta-arrestins.%20%26lt%3Bi%26gt%3BNat%20Commun%26lt%3B%5C%2Fi%26gt%3B%20%26lt%3Bb%26gt%3B17%26lt%3B%5C%2Fb%26gt%3B%2C%209652%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Structural%20basis%20of%20active%20state%20coupling%20of%20metabotropic%20glutamate%20receptor%208%20to%20beta-arrestins%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dagan%20C.%22%2C%22lastName%22%3A%22Marx%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kevin%22%2C%22lastName%22%3A%22Huynh%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alberto%20J.%22%2C%22lastName%22%3A%22Gonzalez-Hernandez%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexa%22%2C%22lastName%22%3A%22Strauss%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Carlos%22%2C%22lastName%22%3A%22Rico%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Pamela%20N.%22%2C%22lastName%22%3A%22Gallo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sheida%22%2C%22lastName%22%3A%22Sharghi%20Moshtaghin%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Anisul%22%2C%22lastName%22%3A%22Arefin%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Johannes%22%2C%22lastName%22%3A%22Broichhagen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22David%22%2C%22lastName%22%3A%22Eliezer%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22George%22%2C%22lastName%22%3A%22Khelashvili%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Joshua%22%2C%22lastName%22%3A%22Levitz%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222026-09-10%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1038%5C%2Fs41467-026-77556-3%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fwww.nature.com%5C%2Farticles%5C%2Fs41467-026-77556-3%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%222041-1723%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A20%3A25Z%22%7D%7D%2C%7B%22key%22%3A%226LMYQ7HI%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Xie%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-09%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BXie%2C%20H.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Diamond%20Agent%3A%20Agentic%20Control%20of%20Federated%20HPC%20Resources%20as%20a%20Service.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.06181%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.06181%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Diamond%20Agent%3A%20Agentic%20Control%20of%20Federated%20HPC%20Resources%20as%20a%20Service%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Haotian%22%2C%22lastName%22%3A%22Xie%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Junlin%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mingkai%22%2C%22lastName%22%3A%22Zheng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yifan%22%2C%22lastName%22%3A%22Zhu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Minu%22%2C%22lastName%22%3A%22Mathew%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Max%22%2C%22lastName%22%3A%22Burnette%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yadu%22%2C%22lastName%22%3A%22Babuji%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Volodymyr%22%2C%22lastName%22%3A%22Kindratenko%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shivaram%22%2C%22lastName%22%3A%22Venkataraman%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kyle%22%2C%22lastName%22%3A%22Chard%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ian%22%2C%22lastName%22%3A%22Foster%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhao%22%2C%22lastName%22%3A%22Zhang%22%7D%5D%2C%22abstractNote%22%3A%22Efficiently%20aggregating%20and%20orchestrating%20computing%20power%20across%20heterogeneous%20clusters%20for%20HPC%20workflows%20faces%20four%20practical%20challenges%3A%20preserving%20workflow%20context%20across%20independently%20administered%20clusters%2C%20moving%20large%20datasets%20between%20sites%2C%20reasoning%20about%20site-specific%20environments%20and%20scheduler%20policies%2C%20and%20exploiting%20live%20queue%20and%20resource%20states%20for%20efficient%20task%20scheduling.%20To%20this%20end%2C%20we%20design%20Diamond%20Agent%2C%20an%20agentic%20system%20that%20enables%20intelligent%20execution%20of%20HPC%20workflows%20across%20heterogeneous%20clusters%20with%20typed%20skills%20as%20the%20interface.%20Diamond%20Agent%20provides%20an%20agent-facing%20workspace%20and%20skills%20that%20unify%20cross-site%20resource%20discovery%2C%20resource%20specification%2C%20data%20movement%2C%20task%20execution%2C%20and%20result%20retrieval.%20A%20centralized%20Diamond%20Agent%20instance%20can%20operate%20multiple%20supercomputers%20without%20being%20deployed%20separately%20on%20each%20login%20node.%20Diamond%20Agent%20translates%20high-level%20agent%20actions%20into%20valid%20site-specific%20executions%2C%20moves%20data%20through%20Globus%20Transfer%2C%20and%20uses%20live%20system%20capability%20and%20queue%20information%20to%20select%20feasible%20placements.%20Its%20event-driven%20continuation%20mechanism%20decouples%20agent%20actions%20from%20long-running%20batch%20jobs%3A%20persistent%20services%20monitor%20remote%20execution%20and%20resume%20the%20agent%20only%20when%20a%20result%20or%20decision-relevant%20event%20is%20available.%20We%20experiment%20with%2027%20hours%20of%20telemetry%20and%2019%20matched%20multi-site%20submission%20rounds%20comprising%2083%20jobs%20across%20four%20production%20supercomputers.%20Compared%20with%20a%20fixed-site%20baseline%2C%20Diamond%20Agent%20reduces%20the%20median%20additional%20completion%20time%20relative%20to%20the%20fastest%20observed%20placement%20from%2042%20seconds%20to%204%20seconds%2C%20a%2010.5x%20reduction.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-09T00%3A40%3A29Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.06181%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.06181%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A15%3A48Z%22%7D%7D%2C%7B%22key%22%3A%22WJKAKBMR%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hixson%22%2C%22parsedDate%22%3A%222026%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHixson%2C%20C.%20A.%20UNDERSTANDING%20AND%20DESIGN%20OF%20MOLECULAR%20DYNAMICS%26%23xA0%3B%20BASED%20SIMULATION%20METHODS.%20%28University%20of%20Oklahoma%2C%20Norman%2C%20Oklahoma%2C%202026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22thesis%22%2C%22title%22%3A%22UNDERSTANDING%20AND%20DESIGN%20OF%20MOLECULAR%20DYNAMICS%20%5CnBASED%20SIMULATION%20METHODS%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Christoper%20Adam%22%2C%22lastName%22%3A%22Hixson%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22thesisType%22%3A%22Doctor%20of%20Philosophy%22%2C%22university%22%3A%22University%20of%20Oklahoma%22%2C%22date%22%3A%222026%22%2C%22DOI%22%3A%22%22%2C%22ISBN%22%3A%22%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fshareok.org%5C%2Fitems%5C%2Fd9cea340-c291-4e76-a783-521676d4afaa%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A14%3A27Z%22%7D%7D%2C%7B%22key%22%3A%22DZWAERKZ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Koric%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-07%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BKoric%2C%20A.%20A.%2C%20Liu%2C%20Q.%20%26amp%3B%20Koric%2C%20S.%20Disentangling%20Attention%20in%20Deep%20Operator%20Learning%3A%20A%20Controlled%20Study%20of%20Data-Driven%20and%20Physics-Informed%20Architectures.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.04407%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.04407%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Disentangling%20Attention%20in%20Deep%20Operator%20Learning%3A%20A%20Controlled%20Study%20of%20Data-Driven%20and%20Physics-Informed%20Architectures%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Amar%20Alem%22%2C%22lastName%22%3A%22Koric%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qibang%22%2C%22lastName%22%3A%22Liu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Seid%22%2C%22lastName%22%3A%22Koric%22%7D%5D%2C%22abstractNote%22%3A%22Deep%20neural%20operators%20learn%20mappings%20between%20input%20functions%20and%20complete%20PDE%20solution%20fields%2C%20enabling%20forward%20evaluations%20of%20new%20problem%20instances%20orders%20of%20magnitude%20faster%20than%20conventional%20numerical%20solvers.%20Attention%20mechanisms%20have%20recently%20been%20introduced%20into%20neural%20operators%2C%20but%20most%20studies%20change%20several%20architectural%20components%20at%20once%2C%20making%20it%20difficult%20to%20identify%20what%20actually%20improves%20accuracy.%20This%20work%20presents%20a%20controlled%20and%20systematic%20study%20of%20five%20deep%20operator%20network%20%28DeepONet%29%20variants%20with%20distinct%20attention%20mechanisms%2C%20trained%20under%20both%20data-driven%20and%20physics-informed%20regimes%2C%20to%20isolate%20the%20effects%20of%20cross-attention%2C%20self-attention%2C%20tokenization%2C%20and%20attention%20depth.%20We%20evaluate%20them%20on%20a%20source-driven%20transient%20one-dimensional%20nonlinear%20diffusion-reaction%20equation%2C%20a%20transient%20one-dimensional%20viscous%20Burgers%20equation%20with%20variable%20initial%20conditions%2C%20and%20a%20two-dimensional%20Poisson%20heat-conduction%20problem%20with%20heterogeneous%20source%20fields.%20Per-sensor%20tokenization%20with%20cross-attention%20reduces%20the%20mean%20relative%20L_2%20error%20of%20the%20classical%20DeepONet%20in%20all%20benchmark-training%20combinations%20by%20factors%20of%202.4-28.0%2C%20while%20the%20best%20attention%20configurations%20reach%203.5-32.3.%20Branch%20self-attention%20paired%20only%20with%20dot-product%20fusion%20is%20inconsistent%2C%20degrading%20the%20one-dimensional%20problems%20while%20helping%20the%20more%20complex%20two-dimensional%20source%20field%3B%20added%20on%20top%20of%20cross-attention%20it%20improves%20all%20six%20cases%2C%20though%20by%20less%20than%20cross-attention%20fusion%20alone.%20Global%20pre-mixing%20provides%20no%20consistent%20benefit.%20Increasing%20cross-attention%20depth%20further%20improves%20accuracy%2C%20but%20with%20diminishing%20returns%20and%20a%20substantially%20higher%20cost%20under%20physics-informed%20training.%20Overall%2C%20query-dependent%20cross-attention%20is%20the%20most%20reliable%20mechanism%2C%20whereas%20branch%20self-attention%20is%20most%20useful%20for%20large%2C%20spatially%20complex%20functional%20inputs.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-07T00%3A06%3A03Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.04407%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.04407%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A09%3A02Z%22%7D%7D%2C%7B%22key%22%3A%22BS69X4KH%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Mahajan%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-15%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMahajan%2C%20J.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20ReaDiT%20Guidance%3A%20Control%20for%20Image%20and%20Video%20Generation%20using%20Diffusion%20Transformer%20Features.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.04649%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.04649%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22ReaDiT%20Guidance%3A%20Control%20for%20Image%20and%20Video%20Generation%20using%20Diffusion%20Transformer%20Features%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jay%22%2C%22lastName%22%3A%22Mahajan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chang%22%2C%22lastName%22%3A%22Liu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rauf%22%2C%22lastName%22%3A%22Makharov%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Viraj%22%2C%22lastName%22%3A%22Shah%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexander%22%2C%22lastName%22%3A%22Schwing%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Svetlana%22%2C%22lastName%22%3A%22Lazebnik%22%7D%5D%2C%22abstractNote%22%3A%22We%20present%20DiT%20Readout%20%28ReaDiT%29%20Guidance%2C%20a%20lightweight%20framework%20for%20controlling%20generation%20with%20Diffusion%20Transformer%20%28DiT%29%20models%20via%20their%20internal%20feature%20representations.%20ReaDiT%20Guidance%20uses%20features%20from%20a%20single%20DiT%20block%20to%20steer%20the%20generative%20process%20according%20to%20spatial%20targets%20-%20like%20depth%2C%20pose%2C%20or%20edge%20maps%20-%20provided%20at%20test%20time.%20Furthermore%2C%20since%20modern%20text-to-video%20models%20are%20largely%20built%20on%20DiT%20backbones%2C%20ReaDiT%20Guidance%20naturally%20extends%20to%20video%20generation%2C%20enabling%20camera%20and%20motion%20control.%20Experimental%20results%20demonstrate%20that%20our%20approach%20achieves%20competitive%20or%20improved%20results%20compared%20to%20existing%20feature-based%20and%20off-the-shelf%20adapter-based%20approaches%20while%20requiring%20fewer%20parameters.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-15T00%3A53%3A48Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.04649%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.04649%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A06%3A40Z%22%7D%7D%2C%7B%22key%22%3A%22IWVG8S49%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Ramasubramanian%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-10%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BRamasubramanian%2C%20S.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Tail-Likelihood%20Reinforcement%20Learning.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.02987%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.02987%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Tail-Likelihood%20Reinforcement%20Learning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shrinivas%22%2C%22lastName%22%3A%22Ramasubramanian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Daman%22%2C%22lastName%22%3A%22Arora%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fahim%22%2C%22lastName%22%3A%22Tajwar%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Guanning%22%2C%22lastName%22%3A%22Zeng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qingyang%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhongzhu%22%2C%22lastName%22%3A%22Zhou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chenfeng%22%2C%22lastName%22%3A%22Xu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Haiwen%22%2C%22lastName%22%3A%22Feng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yuda%22%2C%22lastName%22%3A%22Song%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Aarti%22%2C%22lastName%22%3A%22Singh%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ruslan%22%2C%22lastName%22%3A%22Salakhutdinov%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22J.%20Andrew%22%2C%22lastName%22%3A%22Bagnell%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jeff%22%2C%22lastName%22%3A%22Schneider%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Andrea%22%2C%22lastName%22%3A%22Zanette%22%7D%5D%2C%22abstractNote%22%3A%22Reinforcement%20learning%20typically%20optimizes%20average%20reward.%20For%20generative%20policies%2C%20the%20average%20can%20hide%20an%20important%20distinction%3A%20two%20policies%20can%20achieve%20the%20same%20mean%20reward%20while%20having%20very%20different%20chances%20of%20producing%20a%20rare%20but%20high-reward%20rollout.%20This%20matters%20as%20sampling%20increases%20during%20training%20and%20inference%2C%20since%20its%20benefit%20depends%20on%20retaining%20probability%20mass%20on%20high-reward%20outcomes.%20We%20propose%20to%20optimize%20this%20coverage%20directly.%20Rather%20than%20considering%20only%20expected%20reward%2C%20we%20consider%20all%20of%20its%20upper%20tails%3A%20for%20each%20reward%20threshold%2C%20how%20likely%20is%20the%20policy%20to%20exceed%20it%3F%20This%20turns%20a%20continuous%20reward%20into%20a%20family%20of%20binary%20success%20events.%20We%20introduce%20Tail-Likelihood%20Reinforcement%20Learning%20%28TailRL%29%2C%20which%20maximizes%20the%20log-probability%20of%20exceeding%20a%20randomly%20chosen%20reward%20threshold.%20Its%20gradient%20gives%20more%20weight%20to%20rare%2C%20high-reward%20rollouts%20and%20can%20be%20interpreted%20as%20a%20mixture%20of%20Best-of-k%20gradients.%20TailRL%20requires%20only%20a%20simple%20modification%20to%20the%20advantage%20function%2C%20making%20it%20compatible%20with%20existing%20reinforcement%20learning%20pipelines.%20Across%20object%20localization%2C%20maze%20navigation%2C%20GUI%20grounding%2C%20and%20code%20optimization%2C%20TailRL%20leverages%20rare%20high-reward%20training%20samples%20to%20avoid%20suboptimal%20solutions%20and%20yields%20models%20that%20benefit%20more%20from%20additional%20samples%20at%20inference%20time.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-10T01%3A00%3A36Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.02987%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.02987%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T22%3A02%3A40Z%22%7D%7D%2C%7B%22key%22%3A%22ECPS3LYV%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Zheng%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-04%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BZheng%2C%20H.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20SLIDEFORGE%3A%20An%20LLM%20Agent%20for%20Controllable%20Editing%20of%20Slides%20as%20Structured%20Artifacts.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.03109%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.03109%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22SLIDEFORGE%3A%20An%20LLM%20Agent%20for%20Controllable%20Editing%20of%20Slides%20as%20Structured%20Artifacts%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Haozhen%22%2C%22lastName%22%3A%22Zheng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fulin%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tianhu%22%2C%22lastName%22%3A%22Xiong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yingjie%22%2C%22lastName%22%3A%22Yu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shengyi%22%2C%22lastName%22%3A%22Qian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hanchao%22%2C%22lastName%22%3A%22Yu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alex%22%2C%22lastName%22%3A%22Schwing%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Klara%22%2C%22lastName%22%3A%22Nahrstedt%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mingyuan%22%2C%22lastName%22%3A%22Wu%22%7D%5D%2C%22abstractNote%22%3A%22Current%20AI%20agents%20compellingly%20describe%20slides.%20However%2C%20AI-assisted%20slide%20editing%20requires%20more%20than%20understanding%3A%20the%20output%20must%20retain%20layout%2C%20style%2C%20component%20structure%2C%20and%20native%20editability.%20Towards%2C%20AI-assisted%20slide%20editing%2C%20existing%20agents%20operate%20on%20screenshots%20or%20weak%20document%20representations%20and%20often%20fragment%20coherent%20visual%20units%2C%20rasterize%20editable%20content%2C%20or%20break%20layout.%20In%20contrast%2C%20for%20controllable%20slide%20editing%2C%20we%20introduce%20an%20agentic%20framework%2C%20SLIDEFORGE%2C%20which%20builds%20a%20Deck%20State%20Graph%2C%20an%20executable%20slide%20state%20that%20links%20visual%20decomposition%2C%20native%20pptx%20object%20structure%2C%20and%20perceptual%20organization.%20By%20recovering%20human-referable%20components%20while%20retaining%20fine-grained%20editable%20structure%2C%20SLIDEFORGE%20supports%20theme-preserving%20reconstruction%20through%20slide-native%20operations%20and%20rendered-state%20verification.%20We%20further%20introduce%20an%20evaluation%20paradigm%20for%20controllable%20slide%20transformation%20that%20jointly%20measures%20component%20recovery%2C%20preservation%2C%20restyling%20consistency%2C%20visual%20quality%2C%20and%20native%20editability.%20Experiments%20show%20that%20SLIDEFORGE%20outperforms%20direct%20prompting%2C%20screenshot-based%20agents%2C%20and%20generic%20code-agent%20baselines%20across%20these%20dimensions.%20Code%20is%20available%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2FUIUC-MONET%5C%2FSLIDEFORGE.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-04T00%3A07%3A30Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.03109%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.03109%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A57%3A26Z%22%7D%7D%2C%7B%22key%22%3A%22C65U5IVJ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Munir%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-02%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMunir%2C%20M.%20B.%2C%20Ononto%2C%20A.%20J.%2C%20Joynab%2C%20N.%20S.%2C%20Thuraisingham%2C%20B.%20%26amp%3B%20Khan%2C%20L.%20TRIS%3A%20A%20Tri-Layer%20Retrieval%20Integrity%20Sieve%20Against%20Knowledge%20Poisoning.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.00470%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.00470%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22TRIS%3A%20A%20Tri-Layer%20Retrieval%20Integrity%20Sieve%20Against%20Knowledge%20Poisoning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Muhaimin%20Bin%22%2C%22lastName%22%3A%22Munir%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Akib%20Jawad%22%2C%22lastName%22%3A%22Ononto%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nazia%20Shehnaz%22%2C%22lastName%22%3A%22Joynab%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Bhavani%22%2C%22lastName%22%3A%22Thuraisingham%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Latifur%22%2C%22lastName%22%3A%22Khan%22%7D%5D%2C%22abstractNote%22%3A%22Retrieval-Augmented%20Generation%20%28RAG%29%20grounds%20large%20language%20models%20in%20external%20corpora%2C%20but%20implicit%20trust%20in%20retrieved%20documents%20creates%20a%20critical%20attack%20surface%3A%20PoisonedRAG%20shows%20that%20a%20handful%20of%20crafted%20passages%20can%20dominate%20dense%20retrieval%20and%20steer%20generation%20toward%20attacker-chosen%20answers.%20We%20present%20the%20Tri-Layer%20Sieve%2C%20a%20middleware%20defense%20that%20sanitizes%20retrieved%20evidence%20through%20cross-embedding-space%20clustering%20with%20an%20independent%20judge%20model%2C%20structural%20filtering%20of%20trigger-payload%20artifacts%2C%20and%20LLM%20consistency%20verification.%20The%20design%20exploits%20a%20key%20weakness%20of%20retrieval-stage%20poisoning%3A%20a%20single%20document%20must%20satisfy%20one%20embedding%20geometry%2C%20one%20internal%20Trigger-Payload%20structure%2C%20and%20one%20generation%20objective%20-%20rarely%20all%20three%20simultaneously%2C%20a%20fragility%20that%20persists%20even%20against%20an%20adaptive%20attacker%20who%20paraphrases%20around%20it.%20On%20Natural%20Questions%2C%20HotpotQA%2C%20and%20MS-MARCO%20with%20Contriever%20retrieval%20%28k%3D50%29%2C%20the%20Sieve%20reduces%20black-box%20Attack%20Success%20Rate%20from%2067.0%5C%2F87.0%5C%2F64.0%25%20to%203.0%5C%2F14.0%5C%2F4.0%25%2C%20mitigates%20white-box%20HotFlip%20attacks%20from%20~74%25%20to%2027.8%25%20on%20NQ%20with%20Layer%203%20enabled%2C%20and%20drives%20poisoned-document%20MRR%20to%200.000%2C%20while%20restoring%20clean%20accuracy%20from%2013-33%25%20under%20attack%20to%2058-76%25.%20Under%20an%20architecture-aware%20adversary%20who%20paraphrases%20triggers%20to%20evade%20the%20structural%20filter%2C%20enabling%20the%20consistency%20layer%20halves%20adaptive%20ASR%20%2832.0%25%20to%2015.0%25%20on%20NQ%29%20while%20raising%20clean%20accuracy%20by%2018%20points%2C%20at%20an%20added%20latency%20of%20~16-19%20s%5C%2Fquery%20under%20live%20retrieval.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-02T00%3A19%3A25Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.00470%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.00470%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A56%3A27Z%22%7D%7D%2C%7B%22key%22%3A%22IES9LWBI%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hou%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-02%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHou%2C%20J.%2C%20Pitre%2C%20P.%2C%20Fang%2C%20Y.%20%26amp%3B%20Wang%2C%20X.%20EDGE%3A%20Error%20Dependency%20Graph-Guided%20Multi-Error%20Attribution%20in%20Multi-Agent%20LLM%20Systems.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.01360%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2609.01360%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22EDGE%3A%20Error%20Dependency%20Graph-Guided%20Multi-Error%20Attribution%20in%20Multi-Agent%20LLM%20Systems%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jun%22%2C%22lastName%22%3A%22Hou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Priya%22%2C%22lastName%22%3A%22Pitre%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yi%22%2C%22lastName%22%3A%22Fang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Xuan%22%2C%22lastName%22%3A%22Wang%22%7D%5D%2C%22abstractNote%22%3A%22Large%20language%20model%20%28LLM%29%20agent%20failures%20often%20contain%20multiple%20related%20errors%20rather%20than%20a%20single%20mistake.%20Existing%20attribution%20methods%20usually%20identify%20a%20responsible%20agent%2C%20step%2C%20or%20root%20cause%2C%20but%20do%20not%20explicitly%20model%20dependency%20between%20errors.%20We%20introduce%20EDGE%2C%20an%20Error%20Dependency%20Graph-guided%20multi-Error%20attribution%20framework.%20EDGE%20constructs%20an%20error%20dependency%20graph%20from%20observed%20error%20events%20and%20validates%20a%20reliable%20causal%20subset%20through%20counterfactual%20rollout.%20The%20inference%20graph%20guides%20a%20two-stage%20LLM-as-judge%20detector%20for%20error%20attribution%2C%20and%20the%20intervention-validated%20subgraph%20provides%20a%20more%20reliable%20basis%20for%20explanation%20and%20repair%20analysis.%20Experiments%20on%20TRAIL%20and%20MAST%20show%20that%20EDGE%20improves%20category-level%20multi-error%20attribution%20across%20most%20evaluated%20models%20and%20settings.%20Experiments%20with%20adapted%20Who%26amp%3BWhen-style%20prompts%20show%20that%20the%20graph%20helps%20across%20prompting%20strategies.%20These%20results%20suggest%20that%20dependency%20structure%20is%20a%20useful%20diagnostic%20prior%20for%20agent%20failures%20beyond%20isolated%20root-cause%20prediction.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-02T01%3A10%3A08Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2609.01360%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2609.01360%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A54%3A40Z%22%7D%7D%2C%7B%22key%22%3A%22CB5CCS7D%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Jiang%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-17%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BJiang%2C%20F.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20TurnBench%3A%20A%20Multi-Domain%20Benchmark%20for%20Turn-Taking%20Dynamics%20in%20Spoken%20Dialogue.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.25218%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.25218%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22TurnBench%3A%20A%20Multi-Domain%20Benchmark%20for%20Turn-Taking%20Dynamics%20in%20Spoken%20Dialogue%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Freeman%22%2C%22lastName%22%3A%22Jiang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ramon%22%2C%22lastName%22%3A%22Sanabria%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Soham%22%2C%22lastName%22%3A%22Deshmukh%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Bandhav%22%2C%22lastName%22%3A%22Veluri%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Simon%20Michael%20Vuch%22%2C%22lastName%22%3A%22Williams%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Elliott%20K.%22%2C%22lastName%22%3A%22Suen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Garreth%22%2C%22lastName%22%3A%22Lee%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kevin%20Yoonho%22%2C%22lastName%22%3A%22Choi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Takuya%22%2C%22lastName%22%3A%22Umeki%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Riku%22%2C%22lastName%22%3A%22Kubo%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sathvik%22%2C%22lastName%22%3A%22Udupa%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chien-yu%22%2C%22lastName%22%3A%22Huang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shih-Yun%20Shan%22%2C%22lastName%22%3A%22Kuan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhuoyan%22%2C%22lastName%22%3A%22Tao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Satyapriya%22%2C%22lastName%22%3A%22Krishna%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sefik%20Emre%22%2C%22lastName%22%3A%22Eskimez%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yu%22%2C%22lastName%22%3A%22Tsao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hung-yi%22%2C%22lastName%22%3A%22Lee%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Shinji%22%2C%22lastName%22%3A%22Watanabe%22%7D%5D%2C%22abstractNote%22%3A%22Speakers%20in%20natural%20conversation%20take%20turns%20speaking%20and%20listening%2C%20deciding%20in%20real%20time%20when%20to%20take%2C%20hold%2C%20or%20yield%20the%20floor.%20However%2C%20turn-taking%20evaluation%20remains%20limited%20due%20to%20the%20lack%20of%20a%20consistent%2C%20linguistically%20grounded%20evaluation%20protocol%20and%20hand-annotated%20data%20covering%20diverse%20conversation%20types.%20To%20address%20this%2C%20we%20present%20TurnBench%2C%20a%20multi-domain%20benchmark%20that%20pairs%20a%2030-hour%2C%20hand-labeled%20corpus%20of%20dyadic%20human%20conversation%20with%20a%20standardized%20evaluation%20protocol%20for%20end-of-turn%20and%20interruption%20detection.%20We%20set%20conversation%20type%20as%20a%20controllable%20experimental%20variable%2C%20covering%20six%20distinct%20interaction%20styles%2C%20and%20triple-annotate%20each%20conversation.%20Benchmarking%2014%20heterogeneous%20turn-taking%20systems%2C%20we%20find%20end-of-turn%20recall%20stable%20across%20types%2C%20while%20interruption%20false%20positives%20are%20strongly%20type-dependent%20and%20concentrated%20in%20backchannel-dense%20interaction%20styles.%20Although%20in%20smooth%20floor%20transfers%20human%20listeners%20begin%20speaking%20a%20median%20151%20ms%20before%20the%20current%20turn%20ends%2C%20no%20current%20system%20performs%20equivalently%20without%20incurring%20excessive%20false%20positives.%20We%20release%20our%20corpus%2C%20a%20104-hour%20training%20set%2C%20and%20a%20public%20leaderboard%20with%20an%20interactive%20dataset%20viewer%20at%20https%3A%5C%2F%5C%2Fturnbench.sesame.com.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-17T01%3A06%3A50Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.25218%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.25218%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A47%3A41Z%22%7D%7D%2C%7B%22key%22%3A%22FWZ9MMFA%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Lakhssassi%20et%20al.%22%2C%22parsedDate%22%3A%222026-08%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BLakhssassi%2C%20A.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Mask-Guided%20Multi-Task%20Learning%3A%20Real-Time%20RGB%20Prediction%20of%20Plant%20Photosynthetic%20Efficiency%20at%20the%20Edge.%20in%20%26lt%3Bi%26gt%3BICPR%202026%26lt%3B%5C%2Fi%26gt%3B%20%28Lyon%2C%20France%2C%202026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Mask-Guided%20Multi-Task%20Learning%3A%20Real-Time%20RGB%20Prediction%20of%20Plant%20Photosynthetic%20Efficiency%20at%20the%20Edge%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Abdellah%22%2C%22lastName%22%3A%22Lakhssassi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Taminul%22%2C%22lastName%22%3A%22Islam%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Cristiana%20Bernardi%22%2C%22lastName%22%3A%22Rankrape%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Naoufal%22%2C%22lastName%22%3A%22Lakhssassi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Karla%22%2C%22lastName%22%3A%22Gage%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Khaled%22%2C%22lastName%22%3A%22Ahmed%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22proceedingsTitle%22%3A%22ICPR%202026%22%2C%22conferenceName%22%3A%22International%20Conference%20of%20Pattern%20Recognition%22%2C%22date%22%3A%222026-08%22%2C%22eventPlace%22%3A%22%22%2C%22DOI%22%3A%22%22%2C%22ISBN%22%3A%22%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Ficpr2026orgteam.github.io%5C%2FWSpapers%5C%2Fpdfs%5C%2Fpaper_0162.pdf%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A40%3A55Z%22%7D%7D%2C%7B%22key%22%3A%225B8WT5BS%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Elabd%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-25%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BElabd%2C%20A.%2C%20Gross%2C%20E.%2C%20Kobylianskii%2C%20D.%2C%20Li%2C%20R.%20%26amp%3B%20Nachman%2C%20B.%20Differentiable%20Parametric%20Simulation%20and%20Reconstruction%20Models%20in%20Parnassus.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.22873%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.22873%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Differentiable%20Parametric%20Simulation%20and%20Reconstruction%20Models%20in%20Parnassus%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Abdelrahman%22%2C%22lastName%22%3A%22Elabd%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Eilam%22%2C%22lastName%22%3A%22Gross%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dmitrii%22%2C%22lastName%22%3A%22Kobylianskii%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Runze%22%2C%22lastName%22%3A%22Li%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Benjamin%22%2C%22lastName%22%3A%22Nachman%22%7D%5D%2C%22abstractNote%22%3A%22Parnassus%20is%20a%20framework%20for%20fast%20detector%20simulation%20and%20reconstruction%2C%20directly%20mapping%20truth-level%20particles%20onto%20reconstructed%20objects.%20Such%20models%20can%20be%20built%20from%20deep%20generative%20networks%20trained%20on%20paired%20samples%2C%20which%20are%20fit%20automatically%20to%20a%20target%20detector%2C%20or%20from%20parametric%20prescriptions%20of%20the%20kind%20used%20by%20Delphes%2C%20which%20are%20constructed%20by%20hand.%20We%20remove%20this%20asymmetry%20by%20making%20the%20parametric%20models%20fully%20differentiable%20so%20that%20their%20parameters%20can%20be%20fit%20to%20a%20target%20sample%20by%20gradient%20descent.%20We%20demonstrate%20closure%20by%20fitting%20a%20parametric%20model%20to%20samples%20from%20a%20known%20configuration%20of%20itself%2C%20recovering%20the%20generating%20parameters%20and%20characterizing%20the%20degeneracies%20among%20them%2C%20and%20we%20present%20a%20first%20fit%20to%20CMS%20full%20simulation.%20The%20resulting%20models%20are%20interpretable%2C%20inexpensive%2C%20and%20run%20in%20the%20standard%20Parnassus%20pipeline%20which%20is%20fully%20Python%20based%20and%20GPU%20enabled.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-25T01%3A20%3A53Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.22873%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.22873%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A34%3A31Z%22%7D%7D%2C%7B%22key%22%3A%22FRQ8IJRZ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Yu%20and%20Ji%22%2C%22parsedDate%22%3A%222026-10%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BYu%2C%20C.%20%26amp%3B%20Ji%2C%20H.%20Sememe%3A%20Causal%20Modeling%20for%20Task-Aware%20Depth%20Pruning%20of%20Large%20Language%20Models.%20in%20%26lt%3Bi%26gt%3BCOLM%202026%26lt%3B%5C%2Fi%26gt%3B%20%28San%20Francisco%2C%20CA%2C%202026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22conferencePaper%22%2C%22title%22%3A%22Sememe%3A%20Causal%20Modeling%20for%20Task-Aware%20Depth%20Pruning%20of%20Large%20Language%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Charles%22%2C%22lastName%22%3A%22Yu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Henry%22%2C%22lastName%22%3A%22Ji%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22proceedingsTitle%22%3A%22COLM%202026%22%2C%22conferenceName%22%3A%22Conference%20on%20Language%20Modeling%22%2C%22date%22%3A%222026-10%22%2C%22eventPlace%22%3A%22%22%2C%22DOI%22%3A%22%22%2C%22ISBN%22%3A%22%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fopenreview.net%5C%2Fpdf%3Fid%3DUTMigobeqk%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A30%3A05Z%22%7D%7D%2C%7B%22key%22%3A%22QJ5HZN86%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hou%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-25%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHou%2C%20J.%2C%20Fang%2C%20Y.%20%26amp%3B%20Wang%2C%20X.%20Role-Specialized%20Mixture-of-Agents%20with%20Open-Weight%20LLMs%20for%20Clinical%20Prediction.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.22176%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.22176%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Role-Specialized%20Mixture-of-Agents%20with%20Open-Weight%20LLMs%20for%20Clinical%20Prediction%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jun%22%2C%22lastName%22%3A%22Hou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yi%22%2C%22lastName%22%3A%22Fang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Xuan%22%2C%22lastName%22%3A%22Wang%22%7D%5D%2C%22abstractNote%22%3A%22Large%20Language%20Models%20%28LLMs%29%20are%20increasingly%20applied%20to%20clinical%20prediction%20tasks%20such%20as%20in-hospital%20mortality%20and%20readmission%20from%20electronic%20health%20records%20%28EHRs%29.%20Privacy%20and%20compliance%20constraints%20motivate%20systems%20that%20can%20be%20deployed%20locally%2C%20which%20has%20increased%20interest%20in%20open-weight%20multi-agent%20designs.%20However%2C%20most%20medical%20multi-agent%20systems%20are%20evaluated%20as%20a%20single%20block%2C%20leaving%20unclear%20which%20agent%20role%20contributes%20to%20prediction%20and%20whether%20retrieval%20drives%20observed%20gains.%20We%20study%20a%20role-specialized%20Mixture-of-Agents%20%28MoA%29%20that%20combines%20medical%20knowledge%20retrieval%20with%20contrastive%20similar-patient%20reasoning.%20By%20varying%20the%20role%20design%20while%20holding%20the%20retrieval%20setup%20fixed%2C%20we%20localize%20the%20main%20effect%20to%20the%20final%20integrator.%20Pairing%20large%20open-weight%20analysts%20with%20a%20small%20open-weight%20integrator%20matches%20closed-model%20prompting%20on%20F1%20for%20mortality%20prediction%20while%20flagging%20substantially%20more%20true%20high-risk%20patients.%20Mechanism%20analysis%20shows%20the%20role%20assignment%20directly%20yields%20a%20high-recall%20operating%20point%20without%20threshold%20tuning.%20The%20effect%20is%20task-dependent%2C%20with%20smaller%20gains%20for%20readmission%20because%20the%20available%20records%20correlate%20weakly%20with%20this%20longer-horizon%20outcome.%20These%20results%20position%20role%20design%20as%20a%20key%20factor%20in%20privacy-constrained%2C%20training-free%20clinical%20LLM%20prediction.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-25T00%3A45%3A15Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.22176%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.22176%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A05%3A12Z%22%7D%7D%2C%7B%22key%22%3A%22QCBB8BSL%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chan%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-22%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BChan%2C%20J.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20ReLay%3A%20Personalized%20LLM-Generated%20Plain-Language%20Summaries%20for%20Better%20Understanding%2C%20but%20at%20What%20Cost%3F%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2605.00468%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2605.00468%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22ReLay%3A%20Personalized%20LLM-Generated%20Plain-Language%20Summaries%20for%20Better%20Understanding%2C%20but%20at%20What%20Cost%3F%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Joey%22%2C%22lastName%22%3A%22Chan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yikun%22%2C%22lastName%22%3A%22Han%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jingyuan%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Samuel%22%2C%22lastName%22%3A%22Fang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lauren%20D.%22%2C%22lastName%22%3A%22Gryboski%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexandra%22%2C%22lastName%22%3A%22Lee%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sheel%22%2C%22lastName%22%3A%22Tanna%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qingqing%22%2C%22lastName%22%3A%22Zhu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhiyong%22%2C%22lastName%22%3A%22Lu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lucy%20Lu%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yue%22%2C%22lastName%22%3A%22Guo%22%7D%5D%2C%22abstractNote%22%3A%22Plain%20Language%20Summaries%20%28PLS%29%20aim%20to%20make%20research%20accessible%20to%20lay%20readers%2C%20but%20they%20are%20typically%20written%20in%20a%20one-size-fits-all%20style%20that%20ignores%20differences%20in%20readers%26%23039%3B%20information%20needs%20and%20comprehension.%20In%20health%20contexts%2C%20this%20limitation%20is%20particularly%20important%20because%20misunderstanding%20scientific%20information%20can%20affect%20real-world%20decisions.%20Large%20language%20models%20%28LLMs%29%20offer%20new%20opportunities%20for%20personalizing%20PLS%2C%20but%20it%20remains%20unclear%20whether%20personalization%20helps%2C%20which%20strategies%20are%20most%20effective%2C%20and%20how%20to%20balance%20personalization%20with%20safety.%20We%20introduce%20ReLay%2C%20a%20dataset%20of%20300%20participant--PLS%20pairs%20from%2050%20lay%20participants%20in%20both%20static%20%28expert-written%29%20and%20interactive%20%28LLM-personalized%29%20settings.%20ReLay%20includes%20user%20characteristics%2C%20health%20information%20needs%2C%20information-seeking%20behavior%2C%20comprehension%20outcomes%2C%20interaction%20logs%2C%20and%20quality%20ratings.%20We%20use%20ReLay%20to%20evaluate%20five%20LLMs%20across%20two%20personalization%20methods.%20Personalization%20improves%20comprehension%20and%20perceived%20quality%2C%20but%20it%20also%20raises%20the%20risk%20of%20reinforcing%20user%20biases%20and%20introducing%20hallucinations%2C%20revealing%20a%20trade-off%20between%20personalization%20and%20safety.%20These%20findings%20highlight%20the%20need%20for%20personalization%20methods%20that%20are%20both%20effective%20and%20trustworthy%20for%20diverse%20lay%20audiences.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-09-22T02%3A15%3A08Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2605.00468%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2605.00468%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T21%3A00%3A38Z%22%7D%7D%2C%7B%22key%22%3A%22FC5HWZPA%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Howes%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-26%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHowes%2C%20W.%2C%20Ahmed%2C%20F.%20%26amp%3B%20Alam%2C%20S.%20B.%20Low-Latency%20Activation-Regularized%20Sparse%20Neural%20Operators%20with%20Distillation%20Assistance%20Towards%20Real-Time%20Edge-Deployable%20Virtual%20Sensing.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.23987%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.23987%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Low-Latency%20Activation-Regularized%20Sparse%20Neural%20Operators%20with%20Distillation%20Assistance%20Towards%20Real-Time%20Edge-Deployable%20Virtual%20Sensing%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22William%22%2C%22lastName%22%3A%22Howes%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Farid%22%2C%22lastName%22%3A%22Ahmed%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Syed%20Bahauddin%22%2C%22lastName%22%3A%22Alam%22%7D%5D%2C%22abstractNote%22%3A%22Virtual%20sensing%20enables%20digital%20twins%20and%20safety-critical%20systems%20to%20reconstruct%20and%20forecast%20spatial-temporal%20physics%20in%20real%20time.%20However%2C%20conventional%20computational%20and%20data-driven%20methods%20often%20face%20challenges%20in%20generalization%2C%20latency%2C%20and%20energy%20efficiency%20for%20edge%20deployment.%20Neural%20operators%20offer%20a%20promising%20alternative%20but%20remain%20reliant%20on%20power-intensive%20hardware.%20Spiking%20neurons%20and%20neuromorphic%20computing%20can%20improve%20efficiency%2C%20yet%20surrogate-gradient%20training%20and%20multi-step%20spiking%20introduce%20convergence%20and%20latency%20challenges.%20We%20propose%20the%20Sparse-Activation-ReLU%20%28SAR%29%20layer%2C%20a%20single-step%20alternative%20that%20promotes%20activation%20sparsity%20without%20surrogate-gradient%20training%20while%20remaining%20compatible%20with%20event-based%20computing.%20Within%20a%20trunk-based%20NOMAD%20architecture%2C%20SAR%20achieves%20over%20a%20fivefold%20improvement%20in%20the%20combined%20Latency-Error-Energy%20%28LEE%29%20metric%20compared%20with%20Variable%20Spiking%20Neuron%20%28VSN%29%20and%20Leaky%20Integrate-and-Fire%20%28LIF%29%20implementations.%20We%20further%20analyze%20spiking%20entropy%20and%20feature%20usage%20and%20introduce%20synthetic%20knowledge%20distillation%2C%20reducing%20the%20LEE%20score%20by%20more%20than%20twofold.%20Finally%2C%20we%20improve%20VSN%20through%20a%20ReLU-based%20spiking%20loss%20and%20graph-neighbor%20thresholding.%20On%20the%20Heat%20Exchanger%20dataset%2C%20these%20approaches%20reduce%20L2%20error%20by%20more%20than%20twofold%20and%20nearly%20sevenfold%2C%20respectively%2C%20while%20reducing%20spiking%20and%20spatial%20aggregation.%20Overall%2C%20the%20work%20presented%20is%20a%20step%20towards%20energy-efficient%20virtual%20sensing%20by%20providing%20an%20alternative%20framework%20that%20can%20be%20positioned%20towards%20neuromorphic%20or%20other%20edge%20device%20integration%20that%20can%20be%20a%20gold%20standard%20to%20compare%20latency%2C%20energy%2C%20and%20error%20performance%20for%20future%20efficient%20designs%20that%20are%20sparsity%20or%20brain-inspired%20spiking%20based.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-26T00%3A21%3A51Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.23987%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.23987%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T20%3A57%3A47Z%22%7D%7D%2C%7B%22key%22%3A%22WL6UD4TZ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Abdelghany%20et%20al.%22%2C%22parsedDate%22%3A%222026-09-18%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BAbdelghany%2C%20M.%20A.%2C%20Ngan%2C%20Z.%20%26amp%3B%20Qerimi%2C%20D.%20A%20self-consistent%20hybrid%20global%26%23x2013%3B2D%20model%20for%20SF%206%20%5C%2FAr%20inductively%20coupled%20plasma%20etchers%20with%20neural-network%20surrogate%20acceleration.%20%26lt%3Bi%26gt%3BJ.%20Phys.%20D%3A%20Appl.%20Phys.%26lt%3B%5C%2Fi%26gt%3B%20%26lt%3Bb%26gt%3B59%26lt%3B%5C%2Fb%26gt%3B%2C%20375103%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22A%20self-consistent%20hybrid%20global%5Cu20132D%20model%20for%20SF%206%20%5C%2FAr%20inductively%20coupled%20plasma%20etchers%20with%20neural-network%20surrogate%20acceleration%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Muhammad%20A%22%2C%22lastName%22%3A%22Abdelghany%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zachariah%22%2C%22lastName%22%3A%22Ngan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dren%22%2C%22lastName%22%3A%22Qerimi%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20We%20present%20a%20self-consistent%20hybrid%20global%5Cu2013two-dimensional%20%282D%29%20model%20for%20reactor-scale%20simulation%20of%20SF%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%206%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5C%2FAr%20Inductively%20coupled%20plasma%20%28ICP%29%20etching.%20Unlike%20prior%20reactor-scale%20hybrids%2C%20which%20couple%20two%20spatial%20solvers%2C%20the%20framework%20couples%20a%20zero-dimensional%20%280D%29%20global%20SF%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%206%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5C%2FAr%20chemistry%20solver%20to%20a%202D%20axisymmetric%20electromagnetic%20and%20species-transport%20solver%20through%20a%20shared%20masked-domain%20representation%20and%20a%20two-level%20Picard%20iteration.%20The%202D%20solver%20transports%20the%20nine%20neutrals%20with%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20n%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20e%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20and%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20T%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20e%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%20ion%20densities%20following%20from%20quasi-neutrality%2C%20while%20the%20full%20Lallement%20SF%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%206%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5C%2FAr%20chemistry%20and%20its%20coupled%20surface%20reactions%20are%20retained%20in%20the%200D%20model%2C%20at%20minutes%20of%20wall-clock%20time%20per%20operating%20point%2C%20far%20below%20a%20full%20multidimensional%20solve.%20The%20principal%20advance%20is%20that%20quantities%20ordinarily%20prescribed%20in%20reactor-scale%202D%20simulations%2C%20such%20as%20the%20power-coupling%20efficiency%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu03b7%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%20the%20electron-density%20profile%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20n%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20e%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%28%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20r%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20z%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%29%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%20and%20the%20electron-temperature%20profile%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20T%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20e%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%28%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20r%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20z%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%29%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%20are%20instead%20emergent%20outputs%2C%20so%20that%20the%20operating%20point%20determines%20the%20plasma%20state.%20Benchmarked%20against%20spatially%20resolved%20wafer-plane%20fluorine%20measurements%2C%20the%20model%20reproduces%20the%20absolute%20wafer-center%20fluorine%20density%20to%20within%206%25%5Cu201320%25%20on%20the%20calibration%20composition%2C%20over-predicts%20a%20blind%20composition%20by%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%201.5%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2013%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%202.1%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu00d7%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2C%20and%20under-predicts%20the%20measured%20center-to-edge%20%5BF%5D%20non-uniformity%20by%20roughly%2015%20percentage%20points.%20Both%20absolute-density%20residuals%20lie%20within%20the%20combined%20measurement%20and%20rate-coefficient%20uncertainty.%20A%20neural-network%20surrogate%20reproduces%20the%20wafer-relevant%20atomic-fluorine%20and%20SF%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%206%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20fields%20at%20sub-second%20inference%2C%20an%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20872%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu00d7%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20acceleration%20on%20a%20local%20workstation%20and%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%201750%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu00d7%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20on%20the%20NCSA%20Delta%20HPC%20system%2C%20each%20against%20the%20standalone%20chemistry%5Cu2013transport%20solve%20on%20that%20platform%3B%20a%2021-channel%20extension%20emulating%20the%20reduced%202D%20state%20runs%20at%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2055%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2013%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2091%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu00d7%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20.%20An%20LXCat-based%20electron-kinetics%20analysis%20finds%20the%20dominant%20low-energy%20rates%20distribution-insensitive%20within%20roughly%2020%25%2C%20supporting%20the%20Maxwellian-averaged%20rates%20retained.%20Because%20the%20plasma%20state%20is%20computed%20rather%20than%20fitted%2C%20the%20converged%20model%20serves%20as%20a%20predictive%20instrument%20for%20testing%20physical%20hypotheses%20and%20operating%20scenarios.%20Together%2C%20the%20validated%20model%20and%20its%20surrogate%20form%20the%20predictive%20kernel%20of%20a%20reactor-scale%20SF%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%206%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5C%2FAr%20ICP%20digital%20twin%20for%20near-real-time%20recipe%20development.%22%2C%22date%22%3A%222026-09-18%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1088%5C%2F1361-6463%5C%2Fae9f67%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fiopscience.iop.org%5C%2Farticle%5C%2F10.1088%5C%2F1361-6463%5C%2Fae9f67%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%220022-3727%2C%201361-6463%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-23T20%3A46%3A56Z%22%7D%7D%2C%7B%22key%22%3A%2283MBA4WJ%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Storbacka%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-18%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BStorbacka%2C%20M.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Strangeness%20Transport%20in%20Binary%20Neutron%20Star%20Mergers.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.15527%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.15527%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Strangeness%20Transport%20in%20Binary%20Neutron%20Star%20Mergers%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Melvin%22%2C%22lastName%22%3A%22Storbacka%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jiaxi%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexander%22%2C%22lastName%22%3A%22Haber%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Elias%20R.%22%2C%22lastName%22%3A%22Most%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jacquelyn%22%2C%22lastName%22%3A%22Noronha-Hostler%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mateus%20Reinke%22%2C%22lastName%22%3A%22Pelicer%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nikolas%22%2C%22lastName%22%3A%22Cruz-Camacho%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Veronica%22%2C%22lastName%22%3A%22Dexheimer%22%7D%5D%2C%22abstractNote%22%3A%22The%20presence%20of%20hyperons%20in%20the%20cores%20of%20neutron%20stars%20opens%20fast%20strangeness%20equilibration%20channels%20that%20can%20produce%20bulk-viscous%20dissipation%20during%20binary%20inspiral.%20Because%20these%20reactions%20coexist%20with%20electron%20%24%5Cu03b2%24-equilibration%2C%20tidal%20compression%20can%20drive%20the%20two%20coupled%20chemical%20imbalances%20far%20beyond%20linear%20response.%20We%20construct%20the%20first%20reaction%20network%20that%20self-consistently%20evolves%20the%20electron%20and%20strangeness%20fractions%20with%20a%20four-dimensional%20strangeness-dependent%20chiral%20mean-field%20%28CMF%29%20equation%20of%20state%2C%20including%20nucleonic%20and%20hyperonic%20Urca%20processes%20and%20non-leptonic%20hyperon%20reactions.%20For%20periodic%20density%20perturbations%2C%20representative%20of%20inspiral%20oscillations%2C%20we%20find%20that%20rapid%20strangeness%20conversion%20can%20generate%20a%20large%20%24%5Cu03b2%24-imbalance%2C%20after%20which%20slow%20%24%5Cu03b2%24-equilibration%20bottlenecks%20strangeness%20relaxation.%20Rather%20than%20decaying%20exponentially%2C%20the%20coupled%20system%20consequently%20exhibits%20dynamically%20important%20algebraic%20decay%20in%20a%20far-from-equilibrium%20regime.%20At%20the%20%24%5C%5Crm%20keV%24%20temperatures%20expected%20during%20inspiral%2C%20this%20nonlinear%20response%20produces%20a%20broad%20enhancement%20of%20the%20effective%20bulk%20viscosity%2C%20reaching%20%24%5C%5Csim10%5E%7B31%7D%5C%5C%2C%5C%5Cmathrm%7Bg%5C%5C%2Ccm%5E%7B-1%7D%5C%5C%2Cs%5E%7B-1%7D%7D%24%20for%20%24320%24%20Hz%20oscillations.%20A%20phenomenological%20estimate%20of%20continuous%20inspiral%20dissipation%20yields%20gravitational-wave%20phase%20shifts%20up%20to%20%24%5C%5Csim0.14%24%20rad%20for%20neutron%20stars%20with%20hyperonic%20cores.%20Self-consistent%2C%20far-from-equilibrium%20strangeness%20transport%20may%20therefore%20provide%20a%20dynamical%20probe%20of%20hyperons%20in%20neutron-star%20interiors.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-18T00%3A46%3A18Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.15527%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.15527%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A51%3A54Z%22%7D%7D%2C%7B%22key%22%3A%2222J8WBBR%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Ojonugwa%20Oluwafemi%20et%20al.%22%2C%22parsedDate%22%3A%222026%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BOjonugwa%20Oluwafemi%2C%20Raisa%20Nusrat%20Chowdhury%2C%20Dapiriye%20Briggs%20%26amp%3B%20Md%20Mahmudur%20Rahman.%20Trustworthy%20Medical%20VQA%20Beyond%20Accuracy.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.13140%5C%2FRG.2.2.18698.63681%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.13140%5C%2FRG.2.2.18698.63681%26lt%3B%5C%2Fa%26gt%3B%20%282026%29%20doi%3A10.13140%5C%2FRG.2.2.18698.63681.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Trustworthy%20Medical%20VQA%20Beyond%20Accuracy%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22name%22%3A%22Ojonugwa%20Oluwafemi%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22name%22%3A%22Raisa%20Nusrat%20Chowdhury%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22name%22%3A%22Dapiriye%20Briggs%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22name%22%3A%22Md%20Mahmudur%20Rahman%22%7D%5D%2C%22abstractNote%22%3A%22%22%2C%22date%22%3A%222026%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.13140%5C%2FRG.2.2.18698.63681%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fwww.researchgate.net%5C%2Fdoi%5C%2F10.13140%5C%2FRG.2.2.18698.63681%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%22%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A50%3A39Z%22%7D%7D%2C%7B%22key%22%3A%2257EAW5UU%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Dong%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-17%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BDong%2C%20P.%2C%20Jia%2C%20Y.%2C%20Finn%2C%20C.%20%26amp%3B%20Sadigh%2C%20D.%20Q-Learning%20With%20World%20Models.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2608.17163%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2608.17163%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Q-Learning%20With%20World%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Perry%22%2C%22lastName%22%3A%22Dong%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yueru%22%2C%22lastName%22%3A%22Jia%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chelsea%22%2C%22lastName%22%3A%22Finn%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Dorsa%22%2C%22lastName%22%3A%22Sadigh%22%7D%5D%2C%22abstractNote%22%3A%22Off-policy%20reinforcement%20learning%20%28RL%29%20has%20become%20increasingly%20sample-efficient%2C%20enabling%20applications%20such%20as%20RL%20fine-tuning%20of%20Vision-Language-Action%20models%20into%20reliable%2C%20high-performing%20policies.%20World%20models%20offer%20a%20further%20lever%20for%20sample%20efficiency%2C%20as%20they%20predict%20state%20changes%20rather%20than%20actions%20alone%2C%20but%20their%20success%20has%20largely%20been%20confined%20to%20supervised%20policy%20learning.%20Prior%20model-based%20RL%20methods%20often%20optimize%20the%20policy%20or%20value%20function%20directly%20on%20imagined%20rollouts%2C%20which%20is%20prone%20to%20compounding%20bias%20and%20struggles%20to%20scale%20to%20large%2C%20high-dimensional%20problems%20such%20as%20real-world%20robotics%2C%20a%20problem%20that%20worsens%20with%20task%20horizon%20and%20visual%20complexity.%20In%20this%20work%2C%20we%20instead%20ask%20whether%20we%20can%20leverage%20world%20models%20directly%20on%20top%20of%20standard%20Q-learning%20to%20improve%20performance%2C%20while%20remaining%20trained%20and%20grounded%20in%20the%20real%2C%20online%20setting.%20We%20propose%20QWM%2C%20a%20framework%20that%20leverages%20world%20models%20to%20perform%20test-time%20search%20over%20imagined%20trajectories%20on%20top%20of%20Q-learning%20to%20select%20high-value%20actions%20during%20both%20online%20rollouts%20and%20evaluation.%20Since%20the%20policy%20and%20value%20function%20are%20trained%20only%20on%20real%20transitions%2C%20QWM%20avoids%20compounding%20model%20bias%20while%20still%20gaining%20the%20sample-efficiency%20benefits%20of%20predictive%20search.%20On%20challenging%20manipulation%20benchmarks%20Robomimic%20and%20LIBERO%2C%20QWM%20significantly%20outperforms%20strong%20prior%20state-of-the-art%20methods%20on%20both%20sample%20efficiency%20and%20performance.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2608.17163%22%2C%22date%22%3A%222026-08-17%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2608.17163%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.17163%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A48%3A49Z%22%7D%7D%2C%7B%22key%22%3A%22A7Q6UWYA%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Mullin%22%2C%22parsedDate%22%3A%222026-08-13%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMullin%2C%20L.%20M.%20Testing%20the%20EPYC%20Conjecture%20on%20Real%20Hardware%3A%20MoA-Guided%20Dense%20Matrix%20Multiplication%20on%20NCSA%20Delta%20%28AMD%20EPYC%207763%20Milan%29.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.11533%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.11533%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Testing%20the%20EPYC%20Conjecture%20on%20Real%20Hardware%3A%20MoA-Guided%20Dense%20Matrix%20Multiplication%20on%20NCSA%20Delta%20%28AMD%20EPYC%207763%20Milan%29%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lenore%20M%22%2C%22lastName%22%3A%22Mullin%22%7D%5D%2C%22abstractNote%22%3A%22A%20companion%20empirical%20study%20conjectured%20that%20MoA-guided%20dense%20matrix%20multiplication%20would%20need%20per-CCD%20recalibration%20on%20AMD%20EPYC%20Bergamo%3B%20allocation%20access%20to%20that%20machine%20was%20declined%2C%20and%20this%20paper%20reports%20the%20resulting%20test%20on%20NCSA%20Delta%2C%20an%20architecturally%20related%20chip.%20Its%20central%20claim%3A%20every%20result%20here%20traces%20to%20shape%2C%20not%20architecture%20in%20the%20abstract%20--%20cache%20shape%2C%20access-pattern%20shape%2C%20and%20co-tenancy%20itself%20treated%20as%20a%20shape%20parameter.%20The%20corrected%20block%20size%20M_C%20%3D%20256%20follows%20directly%20from%20Delta%26%23039%3Bs%20real%20512%20KB%20L2%20divided%20by%20the%20data%20type%26%23039%3Bs%20own%20byte%20width%2C%20no%20parameter%20fit%20after%20the%20fact%2C%20and%20outperforms%20the%20M1-Pro-inherited%20M_C%20%3D%2064%20by%2030--59%25%20--%20a%20block%26%23039%3Bs%20shape%20finally%20matching%20the%20cache%20shape%20it%20was%20always%20meant%20to%20occupy.%20The%20same%20reasoning%20extends%20to%20a%20shared%20level%3A%20co-tenancy%20treated%20as%20raw%20capacity%20fails%20by%20an%20order%20of%20magnitude%2C%20while%20co-tenancy%20treated%20as%20a%20shape%20parameter%20--%20capacity%20divided%20by%20concurrent%20consumers%20--%20survives.%20A%20controlled%20NUMA%20experiment%20isolates%20a%20third%20shape%3A%20GEBP%20and%20MoA-pipelined%20compute%20the%20identical%20result%20at%20identical%20achieved%20bandwidth%2C%20yet%20GEBP%20loses%2013.9%25%20of%20its%20throughput%20to%20remote%20memory%20while%20MoA-pipelined%20loses%20only%202.4%25%20--%20not%20a%20speed%20difference%2C%20a%20difference%20in%20the%20shape%20each%20kernel%26%23039%3Bs%20access%20pattern%20traces%20through%20memory.%20The%20headline%20result%20reproduces%20the%20M1%20Pro%26%23039%3Bs%20exact%20three-of-four%20win%20over%20a%20Strassen-GEBP%20hybrid%2C%20a%20pattern%20AWS%20Graviton4%20did%20not%20achieve.%20One%20question%20is%20named%20rather%20than%20resolved%3A%20whether%20Delta%26%23039%3Bs%20larger%20MoA-over-GEBP%20margin%20reflects%20genuine%20speed%20or%20an%20unmatched%20utilization%20fraction.%20Every%20result%20here%20was%20predicted%20before%20it%20was%20measured%2C%20not%20fitted%20after.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-13T00%3A12%3A48Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.11533%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.11533%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A47%3A18Z%22%7D%7D%2C%7B%22key%22%3A%22XKLG65NB%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Harris%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-10%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHarris%2C%20A.%2C%20Corra%2C%20K.%2C%20Chaudhury%2C%20A.%20%26amp%3B%20Hao%2C%20Y.%20Evaluation-Conditioned%20Training%3A%20Teaching%20Models%20to%20Generalize%20to%20Stronger%20Oversight%20Regimes.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2608.10209%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2608.10209%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Evaluation-Conditioned%20Training%3A%20Teaching%20Models%20to%20Generalize%20to%20Stronger%20Oversight%20Regimes%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alec%22%2C%22lastName%22%3A%22Harris%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Kasey%22%2C%22lastName%22%3A%22Corra%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Archie%22%2C%22lastName%22%3A%22Chaudhury%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yixiong%22%2C%22lastName%22%3A%22Hao%22%7D%5D%2C%22abstractNote%22%3A%22Feedback%20signals%20used%20to%20train%20Large%20Language%20Models%20%28LLMs%29%20are%20the%20primary%20driver%20of%20their%20behavior%20and%20our%20main%20lever%20for%20instilling%20alignment%20with%20human%20values%20and%20objectives.%20However%2C%20a%20key%20limitation%20of%20current%20post-training%20methods%20is%20the%20inability%20of%20human%20annotators%20and%20automated%20reward%20functions%20to%20faithfully%20capture%20the%20feedback%20we%20would%20like%20to%20give.%20We%20introduce%20Evaluation-Conditioned%20Training%20%28ECT%29%2C%20a%20post-training%20framework%20that%20uses%20natural%20language%20to%20condition%20each%20training%20sample%20on%20the%20fidelity%20of%20the%20feedback%20we%20provide%20and%20then%20elicits%20the%20desired%20behavior%20by%20conditioning%20the%20LLM%20on%20a%20high-fidelity%20monitor%20in%20deployment.%20ECT%20is%20aimed%20at%20improving%20performance%20under%20imperfect%20feedback%20and%20works%20as%20an%20add-on%20to%20existing%20algorithms%20such%20as%20SFT%20and%20PPO.%20We%20first%20provide%20a%20conceptual%20framework%20for%20ECT%20and%20discuss%20its%20potential%20to%20address%20persistent%20sources%20of%20reward%20mis-specification.%20Then%20we%20motivate%20ECT%20in%20the%20context%20of%20the%20eliciting%20latent%20knowledge%20%28ELK%29%20problem.%20Finally%2C%20we%20evaluate%20ECT%20on%20two%20proof-of-concept%20experiments%3A%20increasing%20even-handedness%20in%20news%20article%20generation%20and%20reducing%20sycophancy%20on%20an%20arithmetic%20task.%20In%20each%20setting%2C%20we%20utilize%20imperfect%20feedback%2C%20rewarding%20bias%20and%20agreement%20with%20the%20user%2C%20respectively.%20In%20both%20settings%2C%20ECT%20improves%20the%20targeted%20behavior%20relative%20to%20direct%20training.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2608.10209%22%2C%22date%22%3A%222026-08-10%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2608.10209%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.10209%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A45%3A30Z%22%7D%7D%2C%7B%22key%22%3A%22SNFNGDRT%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Mullin%22%2C%22parsedDate%22%3A%222026-08-12%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BMullin%2C%20L.%20M.%20Machine%20Shape%20and%20Hierarchical%20Blocking%3A%20A%20Mathematics%20of%20Arrays%20Formalization%2C%20with%20an%20Open%20Problem%20in%20Hierarchical%20Shape%20Occupancy.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.10119%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.10119%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Machine%20Shape%20and%20Hierarchical%20Blocking%3A%20A%20Mathematics%20of%20Arrays%20Formalization%2C%20with%20an%20Open%20Problem%20in%20Hierarchical%20Shape%20Occupancy%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Lenore%20M%22%2C%22lastName%22%3A%22Mullin%22%7D%5D%2C%22abstractNote%22%3A%22A%20companion%20empirical%20study%20found%20that%20dense%20matrix%20multiplication%20block%20sizes%20calibrated%20on%20Apple%20M1%20Pro%20correspond%20to%20two%20cache-t%20formulas%20that%20mispredict%20badly%20on%20a%20dierent%20chip%26%23039%3Bs%20known%20cache%20sizes.%20This%20paper%20formalizes%20the%20question%20that%20nding%20raises.%20We%20extend%20the%20Mathematics%20of%20Arrays%20%28MoA%29%20framework%26%23039%3Bs%20array-shape%20derivation%20operator%20to%20a%20new%20operator%20that%20derives%20a%20hierarchical%2C%20multi-level%20blocking%20and%20prefetch%20schedule%20from%20a%20machine%26%23039%3Bs%20shape%3A%20an%20ordered%20sequence%20of%20cache-level%20capacities%2C%20bandwidths%2C%20and%20occupancy%20fractions.%20This%20operator%20recovers%20the%20calibrated%20values%20on%20every%20one%20of%20three%20real%20machines%20tested%20to%20date%20as%20a%20special%20case%2C%20reducing%20each%20machine%26%23039%3Bs%20unknowns%20to%20a%20small%20number%20of%20level-specic%20occupancy%20fractions.%20We%20then%20state%20precisely%2C%20without%20claiming%20to%20resolve%2C%20the%20paper%26%23039%3Bs%20central%20open%20problem%3A%20whether%20those%20fractions%20are%20derivable%20from%20more%20primitive%20properties%20co-tenancy%2C%20private-cache-level%20count%2C%20associativity%2C%20prefetcher%20behavior%20or%20are%20fundamentally%20per-architecture%20constants.%20Four%20falsiable%20hypotheses%20are%20stated%20and%20tested%20against%20real%20hardware%2C%20with%20mixed%20results.%20We%20further%20state%20two%20limits%20of%20the%20framework%20explicitly%3A%20it%20requires%20dedicated%2C%20non-virtualized%20hardware%20access%20to%20be%20well-dened%20at%20all%2C%20and%20it%20extends%20only%20partway%20to%20a%20distributed-memory%20network%2C%20where%20realizing%20a%20tile%20across%20nodes%20requires%20a%20separate%20choice%20of%20communication%20algorithm%20the%20framework%20does%20not%20itself%20make.%20A%20rst%2C%20honest%20attempt%20at%20extending%20the%20framework%20toward%20predicting%20throughput%20directly%2C%20not%20just%20block%20size%2C%20closes%20the%20paper%3A%20two%20terms%20prove%20derivable%20from%20a%20specication%20sheet%2C%20one%20requires%20a%20single%20measurement%2C%20and%20one%20tested%20across%20three%20machines%20does%20not%20yet%20transfer%20between%20them.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-12T00%3A06%3A02Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.10119%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.10119%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A44%3A26Z%22%7D%7D%2C%7B%22key%22%3A%22DTZ9EG2W%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Hu%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-11%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BHu%2C%20Y.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20Plasmoid-Mediated%202D%20Magnetic%20Reconnection%20in%20Partially%20Ionized%20Plasmas.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.08448%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.08448%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Plasmoid-Mediated%202D%20Magnetic%20Reconnection%20in%20Partially%20Ionized%20Plasmas%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yue%22%2C%22lastName%22%3A%22Hu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Siyao%22%2C%22lastName%22%3A%22Xu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Grzegorz%22%2C%22lastName%22%3A%22Kowal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22James%20M.%22%2C%22lastName%22%3A%22Stone%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alex%22%2C%22lastName%22%3A%22Lazarian%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Hui%22%2C%22lastName%22%3A%22Li%22%7D%5D%2C%22abstractNote%22%3A%22Magnetic%20reconnection%20in%20partially%20ionized%20plasmas%20is%20an%20important%20channel%20for%20energy%20release.%20While%20the%20plasmoid%20instability%20is%20well%20characterized%20in%202D%20fully%20ionized%20plasmas%2C%20its%20behavior%20in%20the%20presence%20of%20neutral-dominated%20plasma%20remains%20poorly%20understood%20in%20the%20nonlinear%2C%20high-Lundquist-number%20%28%24S%20%3D%2010%5E5%24%29%20regime.%20We%20present%20high-resolution%20%28%2416384%20%5C%5Ctimes%204096%24%20cells%29%20two-dimensional%20two-fluid%20%28ion%20%24%2B%24%20neutral%29%20simulations%20of%20Harris-sheet%20reconnection%20with%20upstream%20plasma%20beta%20%24%5Cu03b2%3D%202%24%2C%20comparing%20fully%20ionized%20and%20partially%20ionized%20%28ionization%20fraction%20%24%5Cu03be%3D%2010%5E%7B-1%7D%24%20and%20%2410%5E%7B-2%7D%24%29%20regimes.%20Neutral-ion%20decoupling%20accelerates%20the%20linear%20tearing%20stage%20and%20alters%20the%20plasmoid%20hierarchy%3A%20the%20large-scale%20%60%60monster%26%23039%3B%26%23039%3B%20plasmoid%20that%20dominates%20the%20fully%20ionized%20case%20is%20suppressed%2C%20and%20the%20sheet%20instead%20fragments%20into%20a%20dense%20chain%20of%20sub-scale%20plasmoids.%20Below%20the%20neutral-ion%20decoupling%20scale%20%24%5C%5Cell_%7B%5C%5Crm%20dec%7D%24%2C%20ions%20concentrate%20into%20the%20plasmoids%2C%20reaching%20peak%20overdensities%20%24%5Cu03c1_i%5C%2F%5Cu03c1_%7Bi%2C0%7D%20%5C%5Capprox%2010%24%20%28%24%5Cu03be%3D%2010%5E%7B-1%7D%24%29%20and%20%243-5%5C%5Ctimes10%5E%7B3%7D%24%20%28%24%5Cu03be%3D%2010%5E%7B-2%7D%24%29%2C%20while%20the%20neutrals%20remain%20comparatively%20smooth.%20This%20local%20pile-up%20raises%20the%20ionization%20fraction%20and%20recouples%20the%20two%20fluids%20within%20the%20plasmoids.%20Measured%20from%20the%20out-of-plane%20electric%20field%20at%20the%20reconnection%20sites%2C%20the%20reconnection%20rate%20in%20the%20%24%5Cu03be%3D%2010%5E%7B-2%7D%24%20case%20achieves%20%24R_%7B%5C%5Crm%20rec%7D%5C%5Capprox0.01%24%2C%20whereas%20the%20%24%5Cu03be%3D%2010%5E%7B-1%7D%24%20case%20rises%20to%20a%20rate%20%24%5C%5Capprox0.02%24%20and%20further%20%240.035%24%20when%20apparent%20coalescence%20occurs.%20In%20the%20%24%5Cu03be%3D%2010%5E%7B-2%7D%24%20case%2C%20the%20ambipolar%20drift%20drives%20a%20rapid%20ion%20inflow%20%24%5C%5Csim0.5%5C%5C%2Cv_%7BA%2C0%7D%24%20into%20the%20layer%20at%20the%20same%20reconnection%20sites%2C%20far%20above%20the%20neutral%20inflow%20velocity%20%24%5C%5Csim0.1%5C%5C%2Cv_%7BA%2C0%7D%24.%20Here%2C%20%24v_%7BA%2C0%7D%24%20is%20the%20upstream%20total%20Alfv%5Cu00e9n%20speed.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-11T00%3A52%3A15Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.08448%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.08448%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A42%3A42Z%22%7D%7D%2C%7B%22key%22%3A%224Y5LE43X%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Jafari%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-07%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BJafari%2C%20S.%2C%20Bilwal%2C%20M.%2C%20Zhou%2C%20F.%2C%20Gelder%2C%20B.%20%26amp%3B%20Jannesari%2C%20A.%20APQF%3A%20Agentic%20Profiling-Guided%20Structured%20Pruning%20and%20Mixed-Precision%20Quantization%20with%20Adaptive%20Fine-Tuning.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.05499%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.05499%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22APQF%3A%20Agentic%20Profiling-Guided%20Structured%20Pruning%20and%20Mixed-Precision%20Quantization%20with%20Adaptive%20Fine-Tuning%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Sadegh%22%2C%22lastName%22%3A%22Jafari%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Mohiuddin%22%2C%22lastName%22%3A%22Bilwal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Fan%22%2C%22lastName%22%3A%22Zhou%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Brian%22%2C%22lastName%22%3A%22Gelder%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ali%22%2C%22lastName%22%3A%22Jannesari%22%7D%5D%2C%22abstractNote%22%3A%22Modern%20deep%20neural%20networks%20achieve%20strong%20performance%2C%20but%20their%20scale%20makes%20them%20costly%20and%20slow%2C%20especially%20on%20resource-constrained%20edge%20devices.%20Pruning%20and%20quantization%20address%20this%2C%20but%20rely%20on%20manual%2C%20expert%20choices%20and%20on%20algorithms%20that%20are%20hard%20to%20apply%20across%20architectures.%20Uniform%20settings%20also%20ignore%20how%20differently%20individual%20layers%20respond%20to%20compression%2C%20which%20costs%20accuracy.%20We%20introduce%20APQF%2C%20an%20agentic%20profiling-guided%20framework%20that%20combines%20structured%20pruning%2C%20mixed-precision%20quantization-aware%20training%2C%20and%20accuracy%20recovery%20in%20one%20automated%20pipeline.%20A%20profiling%20agent%20measures%20how%20cost%20is%20distributed%20across%20the%20model%20and%20how%20sensitive%20each%20part%20is%20to%20pruning%2C%20and%20this%20evidence%20drives%20per-layer%20pruning%20ratios%2C%20per-layer%20bit-widths%2C%20and%20the%20recovery%20strategy%2C%20all%20proposed%20by%20LLM%20planners%20and%20validated%20before%20execution.%20To%20our%20knowledge%2C%20APQF%20is%20the%20first%20framework%20to%20combine%20LLM-guided%2C%20profiling-grounded%20decisions%20with%20a%20fully%20training-aware%20pruning%20and%20quantization%20pipeline%20for%20both%20CNNs%20and%20vision%20transformers.%20We%20evaluate%20APQF%20on%20ResNet%2C%20VGG7%2C%20ViT%2C%20DeiT%2C%20and%20Swin%20using%20ImageNet-1k%20and%20CIFAR-10.%20On%20ImageNet%20it%20cuts%20compute%20to%205.6-7.7%20percent%20of%20the%20original%20bit-operations%2C%20a%2013-18x%20reduction%2C%20while%20keeping%20accuracy%20close%20to%20the%20baseline%2C%20and%20under%20a%20200K-image%20budget%20it%20stays%20roughly%2017%20points%20higher%20in%20Top-1%20than%20existing%20joint%20pruning%20and%20quantization%20methods.%20On%20CIFAR-10%20it%20compresses%20further%20than%20that%20method%20on%20four%20of%20five%20architectures.%20On%20VGG7%20it%20reaches%2093.15%20percent%20using%20only%200.41%20percent%20of%20baseline%20bit-operations%2C%20the%20only%20method%20at%20that%20compression%20level%20to%20improve%20on%20its%20full-precision%20baseline.%20Ablations%20show%20that%20uniform%20compression%20loses%20the%20most%20accuracy%20at%20matched%20compute%2C%20and%20that%20withholding%20profiling%20data%20from%20the%20planner%20hurts%20every%20model.%20Six%20LLM%20planners%2C%20including%20free%20open-weight%20ones%2C%20all%20reach%2097.4-97.9%20percent%20on%20Swin-Tiny.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-07T00%3A15%3A31Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.05499%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.05499%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A33%3A53Z%22%7D%7D%2C%7B%22key%22%3A%22MV5Z83YV%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Sathua%20et%20al.%22%2C%22parsedDate%22%3A%222026-08-07%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BSathua%2C%20R.%2C%20Chen%2C%20H.%20%26amp%3B%20Driggs-Campbell%2C%20K.%20ARGUS%3A%20Aligning%20Robot%20Scene%20Geometry%20Under%20Shifting%20Views%20with%20Large%203D%20Vision%20Models.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.05579%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2608.05579%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22ARGUS%3A%20Aligning%20Robot%20Scene%20Geometry%20Under%20Shifting%20Views%20with%20Large%203D%20Vision%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rishik%22%2C%22lastName%22%3A%22Sathua%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Haonan%22%2C%22lastName%22%3A%22Chen%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Katherine%22%2C%22lastName%22%3A%22Driggs-Campbell%22%7D%5D%2C%22abstractNote%22%3A%22Large-scale%20visuomotor%20policies%20have%20demonstrated%20impressive%20performance%20across%20a%20wide%20range%20of%20robot%20manipulation%20tasks.%20However%2C%20despite%20this%20success%2C%20manipulation%20polices%20often%20entangle%20scene%20geometry%20with%20the%20corresponding%20viewpoint%2C%20learning%20where%20objects%20lie%20in%20an%20image%20rather%20than%20where%20it%20lies%20in%20the%20task%20space.%20This%20entanglement%20inherently%20limits%20the%20corresponding%20policy%26%23039%3Bs%20ability%20to%20learn%20from%20viewpoint-diverse%20datasets%20%28ex.%20DROID%2C%20BridgeV2%29%20and%20generalize%20beyond%20the%20viewpoints%20captured%20in%20their%20training%20data.%20In%20this%20work%2C%20we%20present%20ARGUS%2C%20an%20observation%20pre-processing%20pipeline%20that%20uses%20large-scale%203D%20vision%20models%20to%20align%20image%20observations%20from%20arbitrary%20camera%20viewpoints%20into%20a%20canonical%20viewpoint%20before%20passing%20it%20to%20downstream%20visuomotor%20policies.%20Experiments%20across%20training%20datasets%20with%20varying%20levels%20of%20viewpoint%20diversity%2C%20from%20fixed%20multi-view%20camera%20configurations%20to%20highly%20varied%20camera%20placements%2C%20show%20that%20our%20method%20consistently%20outperforms%20prior%20approaches%20across%20both%20limited-view%20and%20view-diverse%20training%20regimes.%20In%20efficiency%20comparisons%2C%20ARGUS%20demonstrates%20an%20ability%20to%20learn%20from%20view-diverse%20data%2C%20converging%20to%20high%20success%20rates%204-6x%20faster%20than%20previous%20methods%20by%20leveraging%20a%20simplified%20observation%20space.%20Overall%2C%20our%20findings%20show%20that%20leveraging%20large-scale%203D%20vision%20models%20reduces%20the%20learning%20burden%20on%20visuomotor%20policies%2C%20enabling%20more%20efficient%20learning%20from%20large-scale%2C%20viewpoint-diverse%20robot%20datasets.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-08-07T00%3A22%3A43Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2608.05579%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2608.05579%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A32%3A42Z%22%7D%7D%2C%7B%22key%22%3A%222A9FJMFG%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Zhang%20and%20Luo%22%2C%22parsedDate%22%3A%222026-08-16%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BZhang%2C%20Y.%20%26amp%3B%20Luo%2C%20C.%20Gravitational%20Stability%20of%20Hydrated%20SiO%26lt%3Bsub%26gt%3B2%26lt%3B%5C%2Fsub%26gt%3B%20and%20Basaltic%20Crust%20in%20the%20Lower%20Mantle.%20%26lt%3Bi%26gt%3BGeophysical%20Research%20Letters%26lt%3B%5C%2Fi%26gt%3B%20%26lt%3Bb%26gt%3B53%26lt%3B%5C%2Fb%26gt%3B%2C%20e2025GL121232%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Gravitational%20Stability%20of%20Hydrated%20SiO%3Csub%3E2%3C%5C%2Fsub%3E%20and%20Basaltic%20Crust%20in%20the%20Lower%20Mantle%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yanyao%22%2C%22lastName%22%3A%22Zhang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chenxing%22%2C%22lastName%22%3A%22Luo%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20We%20determined%20thermal%20equations%20of%20state%20of%20hydrated%20stishovite%20and%20CaCl%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2010type%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20up%20to%20100%5Cu00a0GPa%20and%203%2C000%5Cu00a0K%20to%20assess%20the%20gravitational%20stability%20of%20hydrated%20stishovite%20and%20basaltic%20crust%20in%20the%20lower%20mantle.%20Using%20machine%5Cu2010learning%20based%20atomistic%20simulations%2C%20we%20examined%20three%20hydrogen%20dissolution%20mechanisms%3A%20hydrogarnet%5Cu2010type%2C%20interstitial%20H%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20O%2C%20and%20coupled%20Al%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%203%2B%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu00a0%2B%5Cu00a0H%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2B%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20substitution.%20Hydrogen%20dissolution%20lowers%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20densities%20throughout%20the%20lower%20mantle%20and%20reduces%20bulk%20sound%20velocities%20at%20its%20upper%20part.%20Hydrated%20basaltic%20crust%20with%201%5Cu20134%5Cu00a0wt%25%20H%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20O%20in%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20remains%201%25%5Cu20132.4%25%20denser%20than%20the%20mantle%20and%20can%20sink%20to%20the%20lowermost%20mantle%2C%20potentially%20forming%20water%5Cu2010rich%20reservoirs%20influencing%20seismic%20and%20electrical%20anomalies%20at%20depths.%20Stishovite%20with%200.3%5Cu20132.2%5Cu00a0wt%25%20H%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20O%20is%20neutrally%20buoyant%20at%20730%5Cu2013800%5Cu00a0km%2C%20supporting%20its%20role%20as%20a%20source%20of%20low%20shear%5Cu2010wave%20velocity%20layers%20beneath%20North%20America%2C%20Japan%2C%20and%20the%20European%20Alps%20subduction%20regions.%20These%20results%20provide%20new%20insights%20into%20water%20storage%20and%20transport%20in%20the%20deep%20mantle.%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%2C%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Plain%20Language%20Summary%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20Subducting%20oceanic%20crust%20carries%20water%20deep%20into%20Earth%26%23039%3Bs%20interior%2C%20but%20how%20this%20water%20is%20stored%20and%20where%20it%20ends%20up%20remain%20open%20questions.%20In%20this%20study%2C%20we%20investigated%20how%20water%20affects%20the%20physical%20properties%20of%20silica%20%28SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%29%2C%20a%20major%20component%20of%20subducted%20crust%2C%20at%20the%20relevant%20pressures%20and%20temperatures%20of%20the%20lower%20mantle.%20We%20used%20atomistic%20simulations%20based%20on%20machine%20learning%20to%20investigate%20how%20hydrogen%20atoms%20dissolve%20into%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20affects%20densities.%20Our%20results%20show%20that%20adding%20water%20makes%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20lighter%20and%20softer%20at%20high%20pressure.%20However%2C%20when%20water%20is%20incorporated%20into%20oceanic%20crust%2C%20the%20rock%20still%20remains%20denser%20than%20the%20surrounding%20mantle%20and%20can%20continue%20sinking%20toward%20the%20core%5Cu2010mantle%20boundary.%20Accumulated%20hydrous%20materials%20may%20form%20water%5Cu2010rich%20reservoirs%20that%20affect%20seismic%20and%20electrical%20signals.%20At%20shallower%20depths%20%28about%20730%5Cu2013800%5Cu00a0km%29%2C%20hydrated%20stishovite%20becomes%20neutrally%20buoyant%2C%20which%20helps%20explain%20why%20seismic%20shear%20waves%20travel%20more%20slowly%20under%20certain%20subduction%20areas.%20These%20findings%20provide%20new%20insight%20into%20how%20water%20is%20transported%20and%20stored%20deep%20within%20the%20Earth.%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%2C%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Key%20Points%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20Thermal%20equations%20of%20state%20of%20hydrated%20stishovite%20and%20CaCl%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2010type%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20are%20determined%20via%20machine%5Cu2010learning%20based%20atomistic%20simulations%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20Hydrated%20basaltic%20crust%20containing%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20with%201%5Cu20134%5Cu00a0wt%25%20H2O%20can%20gravitationally%20sink%20into%20the%20base%20of%20the%20lower%20mantle%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20Hydrated%20SiO%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20with%200.3%5Cu20132.2%5Cu00a0wt%25%20H%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%202%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20O%20is%20gravitationally%20stable%20at%20730%5Cu2013800%5Cu00a0km%20where%20low%20shear%5Cu2010wave%20velocity%20layers%20have%20been%20seismically%20observed%22%2C%22date%22%3A%222026-08-16%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1029%5C%2F2025GL121232%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fagupubs.onlinelibrary.wiley.com%5C%2Fdoi%5C%2F10.1029%5C%2F2025GL121232%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%220094-8276%2C%201944-8007%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A27%3A41Z%22%7D%7D%2C%7B%22key%22%3A%22PN6L2K78%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Ng%20and%20Most%22%2C%22parsedDate%22%3A%222026-07-31%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BNg%2C%20H.%20H.-Y.%20%26amp%3B%20Most%2C%20E.%20R.%20Spinning%20down%20neutron-star%20merger%20remnants%20with%20the%20Tayler-Spruit%20dynamo%3A%20Global%20simulations%20reveal%20the%20formation%20of%20massive%20disks%20and%20neutron-rich%20ejecta.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2607.28556%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2607.28556%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Spinning%20down%20neutron-star%20merger%20remnants%20with%20the%20Tayler-Spruit%20dynamo%3A%20Global%20simulations%20reveal%20the%20formation%20of%20massive%20disks%20and%20neutron-rich%20ejecta%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Harry%20Ho-Yin%22%2C%22lastName%22%3A%22Ng%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Elias%20R.%22%2C%22lastName%22%3A%22Most%22%7D%5D%2C%22abstractNote%22%3A%22Magnetic-field%20amplification%20and%20angular%20momentum%20%28AM%29%20transport%20critically%20shape%20the%20secular%20evolution%2C%20lifetime%2C%20and%20electromagnetic%20signatures%20of%20binary%20neutron-star%20merger%20remnants.%20While%20the%20magnetorotational%20instability%20can%20operate%20in%20the%20outer%20negative-shear%20regions%20of%20the%20neutron-star%20remnant%20and%20accretion%20disk%2C%20the%20positive-shear%2C%20stably%20stratified%20core%20may%20instead%20be%20susceptible%20to%20the%20Tayler-Spruit%20dynamo.%20We%20present%20the%20first%20global%2C%20long-term%20general-relativistic%20neutrino-radiation%20magnetohydrodynamics%20simulations%20of%20a%20neutron-star%20merger%20remnant%20incorporating%20the%20unresolved%20Tayler-Spruit%20dynamo%20through%20a%20new%20mean-field%20dynamo%20subgrid%20prescription.%20Our%20axisymmetric%20simulations%20starting%20from%20a%20realistic%20merger%20remnant%20show%20that%20the%20Tayler-Spruit%20dynamo%20is%20primarily%20active%20in%20high-latitude%20regions%20of%20the%20remnant%20core.%20The%20resulting%20Maxwell%20stresses%20redistribute%20AM%20on%20a%20spin-down%20timescale%20of%20a%20few%20hundred%20milliseconds%2C%20substantially%20flattening%20the%20core%20rotation%20profile%20and%20transferring%20mass%20and%20AM%20from%20the%20outer%20remnant%20into%20the%20disk.%20This%20produces%20a%20more%20massive%2C%20extended%2C%20and%20strongly%20magnetized%20disk%20with%20a%20low%20electron%20fraction%2C%20leading%20to%20substantially%20more%20neutron-rich%20ejecta.%20Our%20results%20demonstrate%20that%20currently%20unmodelled%20Tayler-Spruit%20dynamo%20action%20can%20qualitatively%20alter%20the%20rotational%20evolution%2C%20collapse%20prospects%2C%20disk%20formation%2C%20and%20multi-messenger%20signatures%20of%20long-lived%20neutron-star%20merger%20remnants.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-07-31T01%3A08%3A26Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2607.28556%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2607.28556%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A20%3A16Z%22%7D%7D%2C%7B%22key%22%3A%22E44IL324%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Xiao%20et%20al.%22%2C%22parsedDate%22%3A%222026-07-31%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BXiao%2C%20Y.%20%26lt%3Bi%26gt%3Bet%20al.%26lt%3B%5C%2Fi%26gt%3B%20ReToken%3A%20One%20Token%20to%20Improve%20Vision-Language%20Models%20for%20Visual%20Retrieval.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2607.28627%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2607.28627%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22ReToken%3A%20One%20Token%20to%20Improve%20Vision-Language%20Models%20for%20Visual%20Retrieval%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yao%22%2C%22lastName%22%3A%22Xiao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Reuben%22%2C%22lastName%22%3A%22Tan%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhen%22%2C%22lastName%22%3A%22Zhu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yuqun%22%2C%22lastName%22%3A%22Wu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jianfeng%22%2C%22lastName%22%3A%22Gao%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Derek%22%2C%22lastName%22%3A%22Hoiem%22%7D%5D%2C%22abstractNote%22%3A%22Long%20visual%20context%20poses%20a%20challenge%20for%20vision-language%20models%3A%20performance%20degrades%20as%20the%20number%20of%20distractors%20grows%2C%20and%20processing%20all%20tokens%20at%20once%20is%20computationally%20infeasible%20under%20GPU%20memory%20constraints.%20We%20present%20ReToken%2C%20a%20single%20learnable%20embedding%20trained%20as%20an%20explicit%20retrieval%20target%20that%20selects%20a%20sparse%20set%20of%20query-relevant%20visual%20tokens%20from%20a%20pre-filled%20visual%20KV%20cache.%20Trained%20on%20only%20a%20small%20image-QA%20dataset%2C%20ReToken%20yields%20consistent%20gains%20across%20image%20and%20video%20benchmarks%3A%20on%20Visual%20Haystacks%20it%20improves%20Qwen3VL-8B%20by%2013.4%20points%20and%20InternVL3.5%20by%2012.4%20points%20%28%26gt%3B20%25%20relative%29%2C%20and%20on%20LVBench%20it%20transfers%20zero-shot%20to%20long%20video%20for%20an%208.0-point%20gain%20with%20Qwen3VL-8B.%20Thanks%20to%20its%20lightweight%20design%2C%20both%20training%20and%20long-video%20inference%20fit%20on%20a%20single%20H100.%20Code%20is%20available%20at%3A%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Favaxiao%5C%2FReToken%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-07-31T01%3A11%3A07Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2607.28627%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2607.28627%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A19%3A07Z%22%7D%7D%2C%7B%22key%22%3A%22MJX4ARZ9%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Chung%20and%20Yunes%22%2C%22parsedDate%22%3A%222026-07-28%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BChung%2C%20A.%20K.-W.%20%26amp%3B%20Yunes%2C%20N.%20Black%20hole%20ringdown%20spectroscopy%20of%20curvature-coupled%20scalar%20and%20pseudoscalar%20fields.%20%26lt%3Bi%26gt%3BPhys.%20Rev.%20D%26lt%3B%5C%2Fi%26gt%3B%20%26lt%3Bb%26gt%3B114%26lt%3B%5C%2Fb%26gt%3B%2C%20L021505%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Black%20hole%20ringdown%20spectroscopy%20of%20curvature-coupled%20scalar%20and%20pseudoscalar%20fields%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Adrian%20Ka-Wai%22%2C%22lastName%22%3A%22Chung%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Nicol%5Cu00e1s%22%2C%22lastName%22%3A%22Yunes%22%7D%5D%2C%22abstractNote%22%3A%22Remnant%20black%20holes%20formed%20in%20compact%20binary%20coalescences%20provide%20a%20clean%20setting%20for%20probing%20additional%20fields%20coupled%20to%20gravity.%20By%20analyzing%20the%20ringdown%20signals%20of%20the%20loudest%20LIGO-Virgo-KAGRA%20binary-black%20hole%20mergers%2C%20we%20search%20for%20massless%20curvature-coupled%20%28pseudo%29scalar%20fields%20arising%20in%20extensions%20to%20general%20relativity.%20We%20find%20no%20evidence%20for%20such%20fields%20and%20place%2090%25%20credible%20upper%20limits%20on%20the%20coupling%20length%20scale%2C%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2113%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2272%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%28%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2025%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cu2013%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%2045%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%29%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20%5Cn%20%20%20%20%20%20%20%20%20%20%20%20%20%20km%2C%20corresponding%20to%20scales%20significantly%20below%20the%20horizons%20of%20the%20remnant%20black%20holes.%20These%20results%20include%20the%20first%20gravitational-wave%20constraint%20on%20an%20axidilaton-type%20coupling.%20By%20isolating%20the%20ringdown%20phase%20and%20performing%20extensive%20robustness%20checks%2C%20our%20analysis%20provides%20a%20complementary%20observational%20probe%20to%20existing%20full-waveform%20constraints.%20As%20the%20gravitational-wave%20catalog%20expands%2C%20black%20hole%20ringdown%20spectroscopy%20will%20enable%20increasingly%20sensitive%20probes%20of%20gravity%20and%20additional%20fields%20using%20astrophysical%20black%20holes.%22%2C%22date%22%3A%222026-7-28%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1103%5C%2F2ctd-mgb6%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Flink.aps.org%5C%2Fdoi%5C%2F10.1103%5C%2F2ctd-mgb6%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%222470-0010%2C%202470-0029%22%2C%22language%22%3A%22en%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A14%3A20Z%22%7D%7D%2C%7B%22key%22%3A%22T5PLHSM5%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Kaushal%20and%20Penmatsa%22%2C%22parsedDate%22%3A%222026-08-01%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BKaushal%2C%20N.%20%26amp%3B%20Penmatsa%2C%20A.%20M.%20N.%20V.%20FragBERTa%3A%20a%20fragment-aware%20molecular%20representation%20model%20with%20sequential%20attachment-based%20fragment%20embeddings.%20%26lt%3Bi%26gt%3BMach.%20Learn.%3A%20Sci.%20Technol.%26lt%3B%5C%2Fi%26gt%3B%20%26lt%3Bb%26gt%3B7%26lt%3B%5C%2Fb%26gt%3B%2C%20045046%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22FragBERTa%3A%20a%20fragment-aware%20molecular%20representation%20model%20with%20sequential%20attachment-based%20fragment%20embeddings%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Neerav%22%2C%22lastName%22%3A%22Kaushal%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Ajay%20M%20N%20V%22%2C%22lastName%22%3A%22Penmatsa%22%7D%5D%2C%22abstractNote%22%3A%22Abstract%5Cn%20%20%20%20%20%20%20%20%20%20%20%20Automated%20exploration%20of%20chemical%20space%20through%20machine%20learning%20has%20become%20a%20central%20strategy%20in%20modern%20drug%20discovery%2C%20where%20the%20ability%20to%20learn%20transferable%20molecular%20representations%20directly%20impacts%20predictive%20modeling%20performance.%20Recent%20advances%20in%20chemical%20language%20models%20have%20primarily%20relied%20on%20atom-level%20string%20encodings%20such%20as%20SMILES%20or%20SELFIES%2C%20treating%20molecules%20as%20sequences%20of%20individual%20atoms%20or%20characters.%20However%2C%20medicinal%20chemistry%20reasoning%20is%20often%20fragment-centric%2C%20emphasizing%20scaffolds%2C%20functional%20groups%2C%20and%20modular%20substructures%20rather%20than%20isolated%20atoms.%20In%20this%20study%2C%20we%20introduce%20FragBERTa%2C%20a%20molecular%20fragment-aware%20transformer-based%20representation%20language%20model%20pretrained%20using%20masked%20language%20modeling%20on%20Sequential%20Attachment-based%20Fragment%20Embedding%20%28SAFE%29%20representations.%20SAFE%20decomposes%20molecules%20into%20chemically%20meaningful%20fragments%20while%20preserving%20attachment%20information%2C%20enabling%20the%20model%20to%20learn%20fragment-level%20structural%20patterns%20within%20a%20sequence%20framework.%20FragBERTa%20is%20pretrained%20on%2011%20million%20unique%20SAFE%20sequences%20extracted%20from%20multiple%20public%20chemical%20databases%20and%20subsequently%20finetuned%20on%20diverse%20MoleculeNet%20benchmarks%20covering%20both%20regression%20%28ESOL%2C%20FreeSolv%2C%20Lipophilicity%2C%20PDBbind%29%20and%20classification%20%28BACE%2C%20BBBP%2C%20HIV%2C%20Tox21%2C%20SIDER%29%20tasks%20under%20both%20random%20and%20scaffold%20splits.%20Our%20evaluation%20shows%20that%20FragBERTa%20achieves%20competitive%20performance%20across%20tasks%2C%20often%20matching%20or%20surpassing%20prior%20models.%20In%20regression%2C%20it%20demonstrates%20strong%20gains%20in%20lipophilicity%20and%20the%20best%20scaffold-split%20performance%20on%20protein-ligand%20binding%20affinity%20within%20our%20comparison.%20In%20classification%2C%20FragBERTa%20attains%20the%20strongest%20reported%20random-split%20ROC-AUC%20results%20on%20BACE%2C%20HIV%2C%20and%20Tox21%20and%20the%20strongest%20reported%20scaffold-split%20ROC-AUC%20performance%20on%20BBBP%20among%20the%20compared%20baselines%2C%20and%20delivers%20robust%20performance%20across%20bioactivity%2C%20permeability%2C%20and%20toxicity%20benchmarks%2C%20indicating%20that%20fragment-based%20representations%20effectively%20capture%20biologically%20driven%20and%20interaction-sensitive%20endpoints.%20Collectively%2C%20these%20findings%20suggest%20that%20fragment-based%20string%20representations%20offer%20advantages%20over%20atom-level%20representations%20such%20as%20SMILES%20and%20SELFIES%20for%20scaffold-sensitive%20and%20interaction-driven%20tasks%2C%20highlighting%20the%20value%20of%20fragment-level%20inductive%20bias%20for%20future%20molecular%20foundation%20models%20in%20drug%20discovery.%22%2C%22date%22%3A%222026-08-01%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1088%5C%2F2632-2153%5C%2Fae911d%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fiopscience.iop.org%5C%2Farticle%5C%2F10.1088%5C%2F2632-2153%5C%2Fae911d%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%222632-2153%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A11%3A35Z%22%7D%7D%2C%7B%22key%22%3A%22J3V8RU6T%22%2C%22library%22%3A%7B%22id%22%3A5005740%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Gladstone%20et%20al.%22%2C%22parsedDate%22%3A%222026-07-31%22%2C%22numChildren%22%3A0%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%20style%3D%26quot%3Bclear%3A%20left%3B%20%26quot%3B%26gt%3B%5Cn%20%20%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-left-margin%26quot%3B%20style%3D%26quot%3Bfloat%3A%20left%3B%20padding-right%3A%200.5em%3B%20text-align%3A%20right%3B%20width%3A%201em%3B%26quot%3B%26gt%3B1.%26lt%3B%5C%2Fdiv%26gt%3B%26lt%3Bdiv%20class%3D%26quot%3Bcsl-right-inline%26quot%3B%20style%3D%26quot%3Bmargin%3A%200%20.4em%200%201.5em%3B%26quot%3B%26gt%3BGladstone%2C%20A.%2C%20Ji%2C%20H.%20%26amp%3B%20Du%2C%20Y.%20Explorative%20Modeling%3A%20Unlocking%20a%20Third%20Pretraining%20Axis%20and%20End-to-End%20Generation.%20Preprint%20at%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2607.27372%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FARXIV.2607.27372%26lt%3B%5C%2Fa%26gt%3B%20%282026%29.%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%20%20%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Explorative%20Modeling%3A%20Unlocking%20a%20Third%20Pretraining%20Axis%20and%20End-to-End%20Generation%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexi%22%2C%22lastName%22%3A%22Gladstone%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Heng%22%2C%22lastName%22%3A%22Ji%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Yilun%22%2C%22lastName%22%3A%22Du%22%7D%5D%2C%22abstractNote%22%3A%22The%20deep%20learning%20revolution%2C%20kicked%20off%20by%20AlexNet%2C%20taught%20us%20that%20end-to-end%20training%20beats%20decomposing%20a%20problem%20into%20hand-designed%20stages.%20Generative%20modeling%2C%20however%2C%20has%20remained%20the%20exception-despite%20generative%20models%20being%20remarkably%20capable%2C%20they%20are%20still%20not%20trained%20end-to-end.%20This%20is%20because%2C%20at%20its%20core%2C%20generative%20modeling%20is%20about%20handling%20distributions%20with%20many%20modes%2C%20and%20existing%20scalable%20approaches%20handle%20this%20the%20same%20way%2C%20by%20factoring%20the%20generation%20procedure%2C%20which%20prevents%20end-to-end%20generation.%20In%20this%20work%2C%20we%20introduce%20Explorative%20Modeling%2C%20a%20new%20paradigm%20that%20instead%20factors%20the%20training%20loop%2C%20exploring%20K%20candidate%20matches%20between%20model%20generations%20and%20data%2C%20and%20training%20on%20the%20best%2C%20so%20predictions%20commit%20to%20modes%20rather%20than%20blurring%20them.%20We%20find%20Explorative%20Models%20%28XMs%29%20useful%20in%20two%20settings.%20First%2C%20increasing%20exploration%20adds%20a%20third%20pretraining%20axis%20beyond%20parameters%20and%20data%20for%20existing%20generative%20models-where%20scaling%20exploration%20monotonically%20improves%20performance%20across%20both%20continuous%20and%20discrete%20domains%20%28images%2C%20video%2C%20and%20language%29.%20Notably%2C%20gains%20from%20exploration%20increase%20with%20scale%2C%20climbing%20from%207%25%20to%2036%25%20as%20data%20scales%20and%20from%2013%25%20to%2023%25%20as%20models%20grow%2C%20with%20efficiency%20gains%20more%20than%20doubling%20at%203x%20the%20compute.%20Concretely%2C%20exploration%20improves%20FLOP%20efficiency%20by%204.1x%2C%20sample%20efficiency%20by%206.2x%2C%20parameter%20efficiency%20by%2047%25%2C%20lifts%20the%20strongest%20of%20image-generation%20recipes%20to%20a%20near-state-of-the-art%201.43%20FID%20on%20ImageNet%20without%20guidance%2C%20enables%20scaling%20how%20end-to-end%20existing%20models%20are%2C%20and%20unlocks%20scaling%20generalization.%20Second%2C%20XMs%20enable%20end-to-end%20reconstructive%20generative%20modeling%2C%20matching%20diffusion%20on%20control%20tasks%20with%2016-256x%20fewer%20inference%20steps.%20Together%2C%20these%20results%20establish%20XMs%20as%20both%20a%20new%20pretraining%20axis%20for%20existing%20generative%20models%20and%20a%20standalone%20end-to-end%20generative%20modeling%20paradigm.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22%22%2C%22date%22%3A%222026-07-31T00%3A04%3A45Z%22%2C%22DOI%22%3A%2210.48550%5C%2FARXIV.2607.27372%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2607.27372%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22XHSH9DGT%22%5D%2C%22dateModified%22%3A%222026-09-22T16%3A11%3A06Z%22%7D%7D%5D%7D
1.
Mohapatra, R., Lancaster, L., Fielding, D., Quataert, E. & Bryan, G. L. How Magnetic Fields Regulate Cooling and Mixing in Turbulent Radiative Mixing Layers. Preprint at https://doi.org/10.48550/ARXIV.2609.29782 (2026).
1.
Underwood, T. et al. Chronologic: Measuring Language Models’ Ability to Represent the Past. Preprint at https://doi.org/10.48550/arXiv.2609.23178 (2026).
1.
Mohammadi, M. Toscani-Fourier Distance on Probability Measures: Wasserstein Control, Topological Equivalence on Model Classes, and Duality. Preprint at https://doi.org/10.48550/arXiv.2609.23163 (2026).
1.
Li, J., Tian, J. & Watanabe, S. Speech Language Models for Full-Meeting Speaker Diarization: Capabilities and Limitations. Preprint at https://doi.org/10.48550/arXiv.2609.23114 (2026).
1.
Gong, J. et al. SAM-V: Geometry-Aware Segment Anything for Multi-View Instance Segmentation. Preprint at https://doi.org/10.48550/arXiv.2609.25490 (2026).
1.
Kashaniyan, M. & Jannesari, A. Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling. Preprint at https://doi.org/10.48550/ARXIV.2609.19499 (2026).
1.
Hategan-Marandiuc, M. et al. Toward Autonomous Radio Follow-up of Multi-messenger Transients with RADAR: From Alert Parsing to Inference and Observation Scheduling. Preprint at https://doi.org/10.48550/arXiv.2609.18233 (2026).
1.
Wu, Y., Jeon, M., Chang, J.-H. R., Tuzel, O. & Tulsiani, S. RayRoPE: Projective Ray Positional Encoding for Multi-view Attention. Preprint at https://doi.org/10.48550/ARXIV.2601.15275 (2026).
1.
Trachu, T., Cornell, S., Chen, W. & Watanabe, S. LACE: Layer-Wise Compression for Dynamic Frame Rate Codecs. Preprint at https://doi.org/10.48550/ARXIV.2609.17509 (2026).
1.
Gu, L., Huang, H.-J., Qadri, M., Kaess, M. & Yuan, W. TouchAnything: Diffusion-Guided 3D Reconstruction from Sparse Robot Touches. Preprint at https://doi.org/10.48550/ARXIV.2604.08945 (2026).
1.
Zhu, Z., Gong, Y., Xiao, Y., Liu, Y. & Hoiem, D. How to Teach Large Multimodal Models New Skills. Preprint at https://doi.org/10.48550/ARXIV.2510.08564 (2025).
1.
Tang, B., Spalding, Z. & Cogan, G. B. Pretraining for Sample-Efficient Neural Interfaces. Preprint at https://doi.org/10.48550/arXiv.2609.13507 (2026).
1.
Trepka, E. An Evaluation of the Two-Dimensional Gabor Model of First-Layer Filters in Convolutional Neural Networks. in ECCV 2026 Workshop (Malmo, Sweden, 2026).
1.
Syfullah, M. K. & Khalil, A. A. LLM-Based Schema-Aware Split Learning for Privacy-Preserving Mental Distress Prediction Across Heterogeneous Surveys. Preprint at https://doi.org/10.48550/ARXIV.2609.15871 (2026).
1.
Oza, P., Fung, V. & Hu, G. Reaction-Condition- and Ensemble-Aware Screening of Dual-Atom ORR Catalysts via Physics-Informed Machine Learning. ACS Catalysis 16, 18174–18182 (2026).
1.
Ge, Y. et al. MindTopo: Can Foundation Models Reason in Topological Space? Preprint at https://doi.org/10.48550/ARXIV.2609.11900 (2026).
1.
Kwark, D. H. et al. Seamless Whole Slide Label-Free Virtual Staining. Preprint at https://doi.org/10.48550/arXiv.2609.10914 (2026).
1.
Güneş, M. H. et al. Signature of Short-Range Order in Static Response of the Three-Dimensional Electron Gas. Preprint at https://doi.org/10.48550/arXiv.2609.10006 (2026).
1.
Marx, D. C. et al. Structural basis of active state coupling of metabotropic glutamate receptor 8 to beta-arrestins. Nat Commun 17, 9652 (2026).
1.
Xie, H. et al. Diamond Agent: Agentic Control of Federated HPC Resources as a Service. Preprint at https://doi.org/10.48550/ARXIV.2609.06181 (2026).
1.
Hixson, C. A. UNDERSTANDING AND DESIGN OF MOLECULAR DYNAMICS BASED SIMULATION METHODS. (University of Oklahoma, Norman, Oklahoma, 2026).
1.
Koric, A. A., Liu, Q. & Koric, S. Disentangling Attention in Deep Operator Learning: A Controlled Study of Data-Driven and Physics-Informed Architectures. Preprint at https://doi.org/10.48550/ARXIV.2609.04407 (2026).
1.
Mahajan, J. et al. ReaDiT Guidance: Control for Image and Video Generation using Diffusion Transformer Features. Preprint at https://doi.org/10.48550/ARXIV.2609.04649 (2026).
1.
Ramasubramanian, S. et al. Tail-Likelihood Reinforcement Learning. Preprint at https://doi.org/10.48550/ARXIV.2609.02987 (2026).
1.
Zheng, H. et al. SLIDEFORGE: An LLM Agent for Controllable Editing of Slides as Structured Artifacts. Preprint at https://doi.org/10.48550/ARXIV.2609.03109 (2026).
1.
Munir, M. B., Ononto, A. J., Joynab, N. S., Thuraisingham, B. & Khan, L. TRIS: A Tri-Layer Retrieval Integrity Sieve Against Knowledge Poisoning. Preprint at https://doi.org/10.48550/ARXIV.2609.00470 (2026).
1.
Hou, J., Pitre, P., Fang, Y. & Wang, X. EDGE: Error Dependency Graph-Guided Multi-Error Attribution in Multi-Agent LLM Systems. Preprint at https://doi.org/10.48550/ARXIV.2609.01360 (2026).
1.
Jiang, F. et al. TurnBench: A Multi-Domain Benchmark for Turn-Taking Dynamics in Spoken Dialogue. Preprint at https://doi.org/10.48550/ARXIV.2608.25218 (2026).
1.
Lakhssassi, A. et al. Mask-Guided Multi-Task Learning: Real-Time RGB Prediction of Plant Photosynthetic Efficiency at the Edge. in ICPR 2026 (Lyon, France, 2026).
1.
Elabd, A., Gross, E., Kobylianskii, D., Li, R. & Nachman, B. Differentiable Parametric Simulation and Reconstruction Models in Parnassus. Preprint at https://doi.org/10.48550/ARXIV.2608.22873 (2026).
1.
Yu, C. & Ji, H. Sememe: Causal Modeling for Task-Aware Depth Pruning of Large Language Models. in COLM 2026 (San Francisco, CA, 2026).
1.
Hou, J., Fang, Y. & Wang, X. Role-Specialized Mixture-of-Agents with Open-Weight LLMs for Clinical Prediction. Preprint at https://doi.org/10.48550/ARXIV.2608.22176 (2026).
1.
Chan, J. et al. ReLay: Personalized LLM-Generated Plain-Language Summaries for Better Understanding, but at What Cost? Preprint at https://doi.org/10.48550/ARXIV.2605.00468 (2026).
1.
Howes, W., Ahmed, F. & Alam, S. B. Low-Latency Activation-Regularized Sparse Neural Operators with Distillation Assistance Towards Real-Time Edge-Deployable Virtual Sensing. Preprint at https://doi.org/10.48550/ARXIV.2608.23987 (2026).
1.
Abdelghany, M. A., Ngan, Z. & Qerimi, D. A self-consistent hybrid global–2D model for SF 6 /Ar inductively coupled plasma etchers with neural-network surrogate acceleration. J. Phys. D: Appl. Phys. 59, 375103 (2026).
1.
Storbacka, M. et al. Strangeness Transport in Binary Neutron Star Mergers. Preprint at https://doi.org/10.48550/ARXIV.2608.15527 (2026).
1.
Ojonugwa Oluwafemi, Raisa Nusrat Chowdhury, Dapiriye Briggs & Md Mahmudur Rahman. Trustworthy Medical VQA Beyond Accuracy. https://doi.org/10.13140/RG.2.2.18698.63681 (2026) doi:10.13140/RG.2.2.18698.63681.
1.
Dong, P., Jia, Y., Finn, C. & Sadigh, D. Q-Learning With World Models. Preprint at https://doi.org/10.48550/arXiv.2608.17163 (2026).
1.
Mullin, L. M. Testing the EPYC Conjecture on Real Hardware: MoA-Guided Dense Matrix Multiplication on NCSA Delta (AMD EPYC 7763 Milan). Preprint at https://doi.org/10.48550/ARXIV.2608.11533 (2026).
1.
Harris, A., Corra, K., Chaudhury, A. & Hao, Y. Evaluation-Conditioned Training: Teaching Models to Generalize to Stronger Oversight Regimes. Preprint at https://doi.org/10.48550/arXiv.2608.10209 (2026).
1.
Mullin, L. M. Machine Shape and Hierarchical Blocking: A Mathematics of Arrays Formalization, with an Open Problem in Hierarchical Shape Occupancy. Preprint at https://doi.org/10.48550/ARXIV.2608.10119 (2026).
1.
Hu, Y. et al. Plasmoid-Mediated 2D Magnetic Reconnection in Partially Ionized Plasmas. Preprint at https://doi.org/10.48550/ARXIV.2608.08448 (2026).
1.
Jafari, S., Bilwal, M., Zhou, F., Gelder, B. & Jannesari, A. APQF: Agentic Profiling-Guided Structured Pruning and Mixed-Precision Quantization with Adaptive Fine-Tuning. Preprint at https://doi.org/10.48550/ARXIV.2608.05499 (2026).
1.
Sathua, R., Chen, H. & Driggs-Campbell, K. ARGUS: Aligning Robot Scene Geometry Under Shifting Views with Large 3D Vision Models. Preprint at https://doi.org/10.48550/ARXIV.2608.05579 (2026).
1.
Zhang, Y. & Luo, C. Gravitational Stability of Hydrated SiO2 and Basaltic Crust in the Lower Mantle. Geophysical Research Letters 53, e2025GL121232 (2026).
1.
Ng, H. H.-Y. & Most, E. R. Spinning down neutron-star merger remnants with the Tayler-Spruit dynamo: Global simulations reveal the formation of massive disks and neutron-rich ejecta. Preprint at https://doi.org/10.48550/ARXIV.2607.28556 (2026).
1.
Xiao, Y. et al. ReToken: One Token to Improve Vision-Language Models for Visual Retrieval. Preprint at https://doi.org/10.48550/ARXIV.2607.28627 (2026).
1.
Chung, A. K.-W. & Yunes, N. Black hole ringdown spectroscopy of curvature-coupled scalar and pseudoscalar fields. Phys. Rev. D 114, L021505 (2026).
1.
Kaushal, N. & Penmatsa, A. M. N. V. FragBERTa: a fragment-aware molecular representation model with sequential attachment-based fragment embeddings. Mach. Learn.: Sci. Technol. 7, 045046 (2026).
1.
Gladstone, A., Ji, H. & Du, Y. Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation. Preprint at https://doi.org/10.48550/ARXIV.2607.27372 (2026).